# p4r4d0xb0x

> P4R4D0XB0X — Systems Engineer, Rust Developer, AI/ML Engineer. Technical docs and blog on low-level systems, infrastructure, AI/ML, and mathematics. English edition; the Korean original is at https://ql.gl/ko/.

Full content in one file: https://ql.gl/en/llms-full.txt

## Pages

- [About](https://ql.gl/en/about/): p4r4d0xb0x — moving between low-level systems, infrastructure, AI/ML and security, writing down what I read and turning repeated work into tools.
- [Projects](https://ql.gl/en/projects/): Tools and pipelines built by p4r4d0xb0x — a CLI product, a content pipeline, a documentation generator, and this site.

## Docs: ai

- [00. AI](https://ql.gl/en/docs/ai/): // INTELLIGENCE — Where the claims and operational boundaries of AI systems get verified
- [01. AI Agent Platform Verification Guide](https://ql.gl/en/docs/ai/23dc8f35/): Before names and demos, verify sources, data boundaries, failure paths, and operational evidence.

## Docs: category

- [AI::](https://ql.gl/en/docs/category/ai/): Where the claims, the evidence, and the operational limits of AI applications and their underlying tech layers get checked.
- [Etc::](https://ql.gl/en/docs/category/etc/): Notes that fall outside the other categories — agile practice, career roadmaps, and personal records.
- [Infra::](https://ql.gl/en/docs/category/infra/): Hands-on notes on deployment automation, cloud (OCI), Linux operations, and SSH networking.
- [Math::](https://ql.gl/en/docs/category/math/): Math notes for understanding probability, calculus, and optimization from the axioms up. Probability: 5 background posts plus 3 on distributions. Calculus: 5 posts. Optimization: 6 background posts plus 7 core posts.
- [Optimization::](https://ql.gl/en/docs/category/optimization/): Optimization notes arranged for re-reading: problem formulation, background mathematics, optimality conditions, algorithms, and expensive optimization.
- [Philosophy::](https://ql.gl/en/docs/category/philosophy/): 공학자가 마주치는 선택, 가정, 책임을 철학적 질문으로 다시 보는 노트
- [Rust::](https://ql.gl/en/docs/category/rust/): Deep dives into the Rust language, a release-note timeline, workspace operations, and build/deploy notes.
- [Security::](https://ql.gl/en/docs/category/security/): Hacking, security, reverse engineering, and vulnerability analysis — technical depth kept practical.
- [Tools::](https://ql.gl/en/docs/category/tools/): Usage and troubleshooting notes for development tools — git, editors, terminals, and the rest.
- [Typescript::](https://ql.gl/en/docs/category/typescript/): Notes on TypeScript syntax and its ecosystem.
- [Writing::](https://ql.gl/en/docs/category/writing/): Rules for professional technical blogging and the standard for sourcing claims.

## Docs: etc

- [Applying Agile 0x0001](https://ql.gl/en/docs/etc/52ef26dc/): Scrum notes reorganized around the official November 2020 revision of the Scrum Guide (Ken Schwaber,
- [Technical Paper Review and Presentation Template](https://ql.gl/en/docs/etc/676b1d95/): A summary results from structuring, not merely reading. Record claims, evidence, and judgment separately.
- [Applying Agile 0x0002](https://ql.gl/en/docs/etc/819ddb39/): Epic: the broad framework
- [There Is No Royal Road for Programmers](https://ql.gl/en/docs/etc/c153ae6c/): A first-person essay refined from notes I wrote to share with teammates in a development meeting. There are no shortcuts, but there is a repeatable path.
- [Rclone usage](https://ql.gl/en/docs/etc/d20d4b5e/): rclone copy -i -P LOCALPATH r2:R2BUCKETNAME/PATHON_R2/

## Docs: infra

- [00. Infra](https://ql.gl/en/docs/infra/): // OPERATIONS — Deployment, cloud, Linux operations
- [Managing Linux SSH Keys](https://ql.gl/en/docs/infra/1e2a360b/): A well-configured ~/.ssh/config file can cleanly resolve most SSH connection problems.
- [Deploying NestJS with Docker](https://ql.gl/en/docs/infra/2d008743/): A Dockerfile that containerizes a NestJS application with a Docker multi-stage build. Separating the build and runtime stages minimizes image size.
- [Adding an OCI VNIC](https://ql.gl/en/docs/infra/366c6a1a/): Simply attaching a VNIC does not make the OS pick it up right away.
- [IMAP Cron-Job Parsing Bot](https://ql.gl/en/docs/infra/3cb995c0/): A Python script that automatically fetches and parses the latest email from an IMAP server. Running it periodically as a cron job automates email reception.
- [SSH Reverse Proxy](https://ql.gl/en/docs/infra/562045ea/): -L Option
- [Automated GitHub Actions Docker Deployment](https://ql.gl/en/docs/infra/cc1289aa/): A workflow that automatically builds a Docker image with GitHub Actions and pushes it to DockerHub.
- [Connecting Without a Wallet](https://ql.gl/en/docs/infra/d73967a2/): Oracle Autonomous Database uses mTLS by default, which authenticates by downloading an electronic wallet file and installing it on the client. Connecting without managing a wallet requires switching to TLS-only mode. I could not find Korean documentation explaining this for the combination of macOS and DBeaver, so I wrote it down here.
- [Linux Swap Creation Script](https://ql.gl/en/docs/infra/eefafc42/): P4R4D0XB0X — Systems Engineer, Rust Developer, AI/ML Engineer. Low-level systems, infrastructure, and AI.

## Docs: intro

- [00. Introduction](https://ql.gl/en/docs/intro/): // INDEX — a technical documentation archive built from personal experience and study

## Docs: math

- [00. Math](https://ql.gl/en/docs/math/): // FOUNDATIONS — a space for rebuilding probability, differentiation, and optimization from axioms through algorithms
- [11. Mean Value Theorem and Taylor Expansion — Imitating Functions with Polynomials](https://ql.gl/en/docs/math/00c3483e/): Calculus 3/5. The real utility of differentiation is "controlling the behavior of an entire function with one derivative." The mean value theorem is that control mechanism, and repeatedly applying it to reconstruct a function as a polynomial gives the Taylor expansion. Approximation theory, optimization, and the proof of the CLT all begin here.
- [14. Delta Function — Mass Concentrated at a Point](https://ql.gl/en/docs/math/0c890f65/): Calculus 6/6. The Dirac delta is not an ordinary function but a linear operation that extracts the value at one point from a test function. Keep the engineering intuition of "an infinitely narrow spike with total mass 1," but handle calculations safely only inside integrals. Background: limits and differentiation, differentiation rules, and multivariable differentiation.
- [06. Gaussian Distribution — The Distribution of Universality](https://ql.gl/en/docs/math/38875735/): Distribution 1/3. Understand the Gaussian (normal) distribution not as a "bell-shaped curve," but as the unique answer reached by three independent derivations. Background: probability spaces, moments and characteristic functions, and the CLT.
- [05. Bayesian Inference and Conjugate Priors — How Distributions Learn](https://ql.gl/en/docs/math/41fa90ca/): Background 5/5. So far, distributions have been treated as "given." This page covers the procedure for updating beliefs about a distribution after observing data (Bayesian inference) and the special structure that gives this update a closed form (conjugacy). This is where the Beta distribution takes the stage as a "probability distribution over probabilities."
- [09. Limits and the Definition of the Derivative — Inventing the Instantaneous Rate of Change](https://ql.gl/en/docs/math/5f390b21/): Calculus 1/5. Differentiation turns the observation that "a curve looks like a line when viewed closely enough" into mathematics. This page rigorously defines differentiation in the language of limits and explains why the second definition, "linear approximation," is more powerful.
- [02. Expectation, Moments, and Characteristic Functions — Numerical Summaries of Distributions](https://ql.gl/en/docs/math/7c5fb25b/): Background 2/5. This page organizes the tools for summarizing the infinite-dimensional object called a distribution with finitely many numbers (moments) and one function (the characteristic function). The characteristic function is the only practical means of defining alpha-stable distributions, so it receives particular attention.
- [03. Gamma and Beta Functions — A Factory for Normalization Constants](https://ql.gl/en/docs/math/7cd951a1/): Background 3/5. A probability density is complete only after a normalization constant makes its total integral equal to 1. Both the Gaussian's $\sqrt{2\pi}$ and the Beta distribution's $B(\alpha,\beta)$ come from the two special functions on this page.
- [07. Beta Distribution — A Probability Distribution over Probabilities](https://ql.gl/en/docs/math/8be72abc/): Distribution 2/3. The Beta distribution lives on $0,1]$ [Gamma and Beta functions, Bayesian inference and conjugacy.
- [15. Restricted Boltzmann Machine — Learning Distributions through Energy](https://ql.gl/en/docs/math/8cb9ca7f/): Probabilistic Model 1/1. An RBM is a generative model that connects observed and latent variables in a bipartite graph and learns the data distribution through an energy function. The key is adjusting parameters to assign low energy to good data and high energy to poor reconstructions. Background: probability spaces, expectations and moments, and differentiation and optimization.
- [04. Law of Large Numbers, Central Limit Theorem, and Stability — Universal Laws of Sums](https://ql.gl/en/docs/math/8e4df9dc/): Background 4/5. The law of large numbers (LLN), central limit theorem (CLT), and stable distributions emerge in sequence from one question: "what shape does the sum of many independent factors take?" This page covers why Gaussians appear everywhere—and why the Gaussian is not the only answer.
- [08. Alpha-Stable Distributions — The Universal Law of Heavy Tails](https://ql.gl/en/docs/math/a13802ad/): Distribution 3/3. Alpha-stable (α-stable, Lévy stable) distributions are a four-parameter family defined by the stability axiom that "their shape does not change under independent sums." They include Gaussian and Cauchy distributions as special cases and provide the CLT limits for infinite-variance data. Background: characteristic functions, stability and the generalized CLT.
- [10. Differentiation Rules and Techniques — Differentiating the World through Composition](https://ql.gl/en/docs/math/a2a82d60/): Calculus 2/5. Every complex function is built from arithmetic operations and compositions of basic functions. A table of basic derivatives plus four combination rules is therefore enough to differentiate any function mechanically. This "mechanical" property is the seed that later becomes automatic differentiation.
- [13. Differentiation, Optimization, and Automatic Differentiation — Machines That Learn from Gradients](https://ql.gl/en/docs/math/b6afca63/): Calculus 5/5. This page assembles the tools from the previous four parts: why the condition "derivative = 0" is the starting point of optimization, why gradient descent works (Taylor expansion), and how deep-learning frameworks calculate gradients without error (automatic differentiation = systematic application of the chain rule).
- [01. Probability Spaces and Random Variables — Where Understanding Distributions Begins](https://ql.gl/en/docs/math/c2fe3bc7/): Background 1/5. Every distribution in this series is ultimately "a measure induced by a random variable defined on a probability space." The goal of this page is to make that sentence completely interpretable.
- [12. Multivariable Differentiation — From Gradients to Jacobians and Hessians](https://ql.gl/en/docs/math/d54e2dbe/): Calculus 4/5. The moment the input becomes a vector, "one slope" is no longer enough. Because differentiation was defined as the "best linear approximation," the extension is natural: the linear approximation is merely promoted from a number to a vector (gradient), then from a vector to a matrix (Jacobian). Every derivative in deep learning is written in the language of this page.
- [00. Optimization](https://ql.gl/en/docs/math/optimization/): // PROBLEMTOALGORITHM — a map for rereading an optimization problem from its definition through algorithm selection
- [04. Numerical Analysis for Optimization — Making Paper Methods Work in Code](https://ql.gl/en/docs/math/optimization/0b9b307a/): // BACKGROUND 4/6 — approximation error, floating-point arithmetic, iterative method, convergence, condition number
- [11. Newton and Quasi-Newton Methods — Using Curvature while Controlling Cost](https://ql.gl/en/docs/math/optimization/0de50a62/): // CORE 5/7 — Newton, quasi-Newton, DFP, BFGS
- [03. Linear Algebra for Optimization — Directions, Curvature, and Linear Systems](https://ql.gl/en/docs/math/optimization/10034fb4/): // BACKGROUND 3/6 — vector, matrix, inner product, linear independence, eigenvalue, positive definiteness, linear systems
- [12. Coordinate Descent and Conjugate Gradient — Coordinates and Conjugate Directions](https://ql.gl/en/docs/math/optimization/1a8c5674/): // CORE 6/7 — coordinate descent, conjugate gradient, preconditioning
- [01. Mathematical Notation — The Minimum Grammar of Optimization](https://ql.gl/en/docs/math/optimization/23a91627/): // BACKGROUND 1/6 — read sets, intervals, vectors, functions, mappings, summations, and norms in a consistent way
- [06. Engineering Modeling — Turning Real Problems into Optimization Problems](https://ql.gl/en/docs/math/optimization/24fe0d46/): // BACKGROUND 6/6 — input-output model, simulation, ODE/PDE, objective definition, constraint definition
- [10. Gradient Descent and Line Search — Separating Direction from Step Length](https://ql.gl/en/docs/math/optimization/49651a07/): // CORE 4/7 — gradient descent, exact line search, backtracking, Armijo, Wolfe, momentum, Nesterov
- [07. Problem Formulation — Variables, Objectives, Constraints, and the Feasible Region](https://ql.gl/en/docs/math/optimization/754ca82b/): // CORE 1/7 — decision variables, objective function, constraints, feasible region
- [02. Calculus for Optimization — From Rates of Change to Second-Order Approximation](https://ql.gl/en/docs/math/optimization/7ab298b6/): // BACKGROUND 2/6 — derivative, partial derivative, gradient, Hessian, Taylor expansion
- [05. Probability and Statistics Background — Uncertain Evaluations and Data-Driven Objectives](https://ql.gl/en/docs/math/optimization/8f51b0f5/): // BACKGROUND 5/6 — random variable, distribution, expectation, variance, sampling, regression
- [13. Expensive Optimization — Optimization when Evaluations Are Costly](https://ql.gl/en/docs/math/optimization/9c46e25f/): // CORE 7/7 — DOE, Latin Hypercube Sampling, surrogate modeling, surrogate-based optimization
- [09. Optimality Conditions — Identifying Where to Stop](https://ql.gl/en/docs/math/optimization/d3cbd57b/): // CORE 3/7 — first-order condition, second-order condition, positive definiteness
- [08. Mathematical Foundations — Gradient, Convexity, Norm, and Decomposition](https://ql.gl/en/docs/math/optimization/d523e7db/): // CORE 2/7 — gradient/Jacobian/Hessian, Taylor approximation, convexity, norm, inner product, orthogonality, matrix decomposition

## Docs: philosophy

- [00. Philosophy](https://ql.gl/en/docs/philosophy/): // ORIENTATION — What do we believe, and what should we build?
- [02. Do we discover problems, or construct them?](https://ql.gl/en/docs/philosophy/0cacd763/): The question: The alarm went off, but what was the problem?
- [15. Design That Leaves Questions](https://ql.gl/en/docs/philosophy/13113957/): The Problem: A Finished Screen Erases the Questions
- [07. What We Can Do Is Not the Same as What We Should Do](https://ql.gl/en/docs/philosophy/2a6c0e0e/): The Problem: The Door Opened by a Single Button
- [03. Whose common sense is it?](https://ql.gl/en/docs/philosophy/38f20b5c/): The question: A screen nobody calls strange
- [Why Do We Need Philosophy?](https://ql.gl/en/docs/philosophy/3e157a58/): // ORIENTATION — What does the maker believe?
- [11. Where Responsibility Lives](https://ql.gl/en/docs/philosophy/55129976/): The Problem: After an Incident, Only the Logs Remain
- [08. When Does Efficiency Become Violence?](https://ql.gl/en/docs/philosophy/5d7979ca/): The Problem: Smaller Numbers and Longer Nights
- [13. What Do I Believe and Build?](https://ql.gl/en/docs/philosophy/6ea516e7/): The Problem: Code Is a Trace of Belief
- [04. Do models describe reality?](https://ql.gl/en/docs/philosophy/82ab4e02/): The question: The graph rose, so why did people leave?
- [14. What Is Good Technology?](https://ql.gl/en/docs/philosophy/886f6a00/): The Problem: Working and Being Good Are Not the Same
- [10. What Does Free Will Have to Do with System Design?](https://ql.gl/en/docs/philosophy/8a6dd399/): The Problem: Who Pressed the Button?
- [05. Does what cannot be measured exist?](https://ql.gl/en/docs/philosophy/beb073ed/): The question: What is missing from the dashboard?
- [06. Does data tell the truth?](https://ql.gl/en/docs/philosophy/c879df0a/): The question: How does a number testify?
- [09. There Is No Perfect System](https://ql.gl/en/docs/philosophy/f582fdc7/): The Problem: The Illusion of a Day Without Failure
- [12. Can Humanity Be Automated?](https://ql.gl/en/docs/philosophy/f690cfa4/): The Problem: The Office Where the Empathy Button Rings

## Docs: rust

- [00. Rust](https://ql.gl/en/docs/rust/): // LOW-LEVEL — Memory-safe systems programming
- [Rust Mocking Objects](https://ql.gl/en/docs/rust/293ec56f/): - Unit tests have to finish fast and must not be affected by their surroundings.
- [Rust Async Traits: The Boundary Between async fn and Poll-Based I/O](https://ql.gl/en/docs/rust/30131fff/): Introduction
- [Rust Separate Build Configuration with Dynamic Linking](https://ql.gl/en/docs/rust/86e4077f/): Setting prefer-dynamic = true in a Rust workspace enables dynamic-library linking for release builds. It is useful for optimizing build speed and binary size.
- [Workspace Tips for Growing Projects, Part 1](https://ql.gl/en/docs/rust/984d1edb/): Introduction
- [Rust WorkSpace Versioning](https://ql.gl/en/docs/rust/c40d68d9/): cargo install cargo-edit
- [Rust Chrono & Chrono-tz Cheatsheet](https://ql.gl/en/docs/rust/df11c3de/): DateTime Conversion Utils
- [Rust Debugging — rust-gdb & rust-lldb (Mac: lldb)](https://ql.gl/en/docs/rust/eb685c80/): What is LLDB?

## Docs: security

- [00. Security & Hacking](https://ql.gl/en/docs/security/): // OFFENSIVE — Technical depth × practicality
- [Decrypting Saved DBeaver Passwords](https://ql.gl/en/docs/security/0aaae33f/): DBeaver stores database connection passwords locally with AES encryption. If you know the key, you can decrypt them.
- [16. Bypassing Path Traversal Filters with Unicode Normalization](https://ql.gl/en/docs/security/0c698fa0/): When a web server or API filters only ../, differences in Unicode normalization (NFC/NFKC) may
- [66. Complete SSRF Guide - Turning a Server Against Itself](https://ql.gl/en/docs/security/1037d430/): TL;DR: Server-Side Request Forgery (SSRF) is a vulnerability that makes a server send requests to arbitrary URLs chosen by an attacker. It can enable access to internal networks, theft of cloud metadata, port scanning, and other attacks.
- [23. Finding N-day Vulnerabilities Through Patch Diffing](https://ql.gl/en/docs/security/10f565b5/): Immediately after a patch is released, the diff is the best clue. The key is comparing binaries or
- [37. Web Cache Poisoning: Turning a CDN into an Attack Surface](https://ql.gl/en/docs/security/113a9222/): Web caches improve performance, but a misconfigured cache becomes a target for cache poisoning. The key is to “produce different responses for the same URL, then plant one of those responses in the cache.”
- [21. Extracting Flags Through Log Leaks in CTFs](https://ql.gl/en/docs/security/173ba3af/): A surprisingly powerful vector in web and reversing CTFs is the log leak. The key is to identify quickly what gets logged and where.
- [46. Deserialization Vulnerabilities — Triggering RCE Through Object Deserialization](https://ql.gl/en/docs/security/18dc1e0c/): What is deserialization?
- [11. Quick Checks for Host Header Attacks](https://ql.gl/en/docs/security/1c99584c/): Web applications often use the Host value to build redirect URLs, absolute links, and email or password-reset links. If the trust boundaries between the proxy, cache, and application do not align, this can lead to Open Redirects, link hijacking, or cache poisoning.
- [27. WAF Bypass Playbook: Small Variations That Disrupt Signatures](https://ql.gl/en/docs/security/1d3a6c87/): WAFs generally block requests using regular expressions/signatures. The key bypass strategy is therefore to “change the form while preserving the function.” The following variations are common in CTFs and real-world testing.
- [35. Building a Hybrid Reversing Routine with Ghidra + Frida](https://ql.gl/en/docs/security/1e3219c7/): Static analysis shows structure, while dynamic analysis shows real behavior. Combining the two tools in CTFs or real-world analysis dramatically increases speed.
- [59. Web Application Fuzzing: Finding Hidden Bugs](https://ql.gl/en/docs/security/300020aa/): Fuzzing is a technique that sends unexpected input to an application to discover vulnerabilities, hidden endpoints, and edge cases. This guide covers practical fuzzing techniques for web security testing and CTF competitions.
- [44. Introduction to Cryptographic Attack Techniques: From Padding Oracles to Bit-Flipping](https://ql.gl/en/docs/security/381c2dc8/): Encrypted cookies and tokens appear often in CTFs. Assuming "it is encrypted, so it must be safe" is a
- [42. Buffer Overflow Basics: The First Step in Binary Exploitation](https://ql.gl/en/docs/security/3947367d/): A buffer overflow is one of the most fundamental vulnerabilities in software security and essential
- [45. CTF Reconnaissance: Mastering OSINT Techniques](https://ql.gl/en/docs/security/3c408bc1/): OSINT (open-source intelligence) is a core skill in CTF competitions, penetration testing, and security
- [36. Padding Oracle: Secrets Revealed by Ciphertext](https://ql.gl/en/docs/security/3f9e4ebb/): A padding oracle is a decryption interface whose external behavior reveals whether the padding of
- [29. Getting a Feel for HTTP Request Smuggling](https://ql.gl/en/docs/security/414d0656/): Request smuggling occurs when a proxy or load balancer and its backend interpret request boundaries differently. A common combination is disagreement over Content-Length and Transfer-Encoding: chunked.
- [48. Getting Started with Dynamic Analysis Using Frida](https://ql.gl/en/docs/security/442c995b/): Dynamic analysis is the technique of observing and modifying a program's behavior while it runs. Frida
- [56. Shellcode Writing Basics — From Assembly to Bytes](https://ql.gl/en/docs/security/4524dfd2/): Shellcode is a raw sequence of machine-code bytes intended to execute while exploiting a vulnerability. It is inserted as a payload in a buffer overflow or ROP chain and is also a recurring CTF subject.
- [09. DNS Rebinding: The Boundary Between Origin and Endpoint](https://ql.gl/en/docs/security/4797605e/): DNS Rebinding is an attack in which an attacker changes DNS answers for a hostname they control so that a browser first receives code from an external server and later connects through the same hostname to an internal-network or loopback service. The key is not hacking DNS itself, but exploiting the gap between a web origin's identity and its actual network endpoint.
- [40. Use-After-Free: Reclaim Freed Memory](https://ql.gl/en/docs/security/4dafab16/): UAF (Use-After-Free) is a vulnerability that occurs when a pointer continues to be used after its heap memory has been freed with free(). It is continually found in modern browsers, kernels, and server software, and it is a recurring CTF topic.
- [41. API Security Testing: A Practical Guide](https://ql.gl/en/docs/security/4e234195/): Modern applications rely heavily on APIs for communication between services, making API security
- [17. Building Reversing Signatures with YARA Rules](https://ql.gl/en/docs/security/50198e06/): YARA is a tool for declaratively matching patterns in binaries and memory dumps. During reversing,
- [22. GraphQL Introspection Abuse Points](https://ql.gl/en/docs/security/5099b6be/): In GraphQL, the schema is the API documentation itself. If introspection is enabled, an attacker can map the entire query surface. It is also a common early clue in CTFs.
- [02. LD_PRELOAD Runtime Hooking at a Glance](https://ql.gl/en/docs/security/53f40dee/): When patching a binary is difficult in the early stages of reverse engineering, LD_PRELOAD is often the fastest bypass. It overrides functions with a preloaded shared library when the dynamic linker resolves symbols.
- [57. Side-Channel Attacks: Silent Hacking](https://ql.gl/en/docs/security/58cff54e/): Even when a cryptographic algorithm is mathematically perfect, its implementation can be breached. Side-Channel Attacks target that exact gap.
- [28. Finding Web Security Issues by Reversing WASM](https://ql.gl/en/docs/security/62fcf0af/): Modern frontends often hide sensitive logic in WebAssembly. WASM runs in the browser, but it is
- [38. Android APK Reverse Engineering](https://ql.gl/en/docs/security/63a6af8b/): With a single APK, you can extract all of an app's secrets. This is a recurring topic in CTF mobile categories.
- [61. WebSocket Security: Hidden Vulnerabilities in Real-Time Communication](https://ql.gl/en/docs/security/6e3ae330/): Unlike HTTP, WebSocket provides persistent, bidirectional communication. It is widely used in modern web applications for chat, games, and real-time dashboards, but its security design is often weak.
- [01. Validating SSRF Quickly with OAST](https://ql.gl/en/docs/security/6f22d9ed/): When you identify a possible SSRF sink, the key is to check immediately whether it produces outbound DNS or HTTP traffic. OAST (Out-of-Band Application Security Testing) works by triggering an external request and observing its trace, leaving evidence even when a WAF or filter blocks the visible response.
- [39. Introduction to Linux Kernel Exploitation: The World of LPE](https://ql.gl/en/docs/security/723f757e/): Kernel exploitation is called the pinnacle of hacking. LPE (Local Privilege Escalation), which obtains root from ordinary user privileges, appears regularly in the pwn category of CTFs and is also a key stage in real APT attacks.
- [13. Notes on OAuth redirect_uri Pitfalls](https://ql.gl/en/docs/security/72c08a55/): Loose validation of redirect_uri in OAuth/OIDC can lead to token theft or account-linking
- [54. Solving CTF Binaries with Radare2](https://ql.gl/en/docs/security/73fd16bc/): Radare2 (r2) is a powerful open-source reverse-engineering framework that has become essential for players tackling CTF binary-exploitation challenges. While Ghidra and IDA Pro offer polished GUIs, radare2's command-line interface and scripting capabilities provide incredible flexibility for fast analysis.
- [52. Prototype Pollution: JavaScript's Hidden Vulnerability](https://ql.gl/en/docs/security/7b457ac5/): Prototype Pollution is an attack technique that changes an application's overall behavior by polluting JavaScript's prototype chain. The name may be unfamiliar, but it appears fairly often in real-world CTFs and bug bounty programs.
- [Review of Hijacking Bitcoin: Routing Attacks on Cryptocurrencies](https://ql.gl/en/docs/security/7b8d25cb/): When Internet routing becomes unstable, the decentralized overlay above it moves with it.
- [26. Identifying Function Boundaries in Stripped Binaries](https://ql.gl/en/docs/security/7db9d623/): Stripped binaries lack function names and debugging information, making control-flow inference essential. When automatic analysis in Ghidra or IDA becomes unreliable, the following routine helps establish the outline quickly.
- [14. Applying the RSA Common-Modulus Vulnerability in CTFs](https://ql.gl/en/docs/security/7e09efe2/): If RSA reuses the same N (modulus) while encrypting with different values of e, the plaintext can
- [49. Reverse Engineering with Ghidra: A Practical Guide](https://ql.gl/en/docs/security/7f9b9628/): Ghidra is an open-source reverse-engineering framework released by the NSA, and it has become an essential tool for security researchers, CTF participants, and malware analysts alike. This guide covers practical techniques for analyzing binaries with Ghidra.
- [51. JWT Security Vulnerabilities: Common Mistakes and Exploitation Techniques](https://ql.gl/en/docs/security/8666e44a/): JWTs (JSON Web Tokens) are everywhere in modern web applications. They are convenient, stateless, and easy to implement—which is exactly why they are often implemented incorrectly. Let's examine JWT vulnerabilities commonly found in CTFs and real applications.
- [64. Reverse Engineering 101](https://ql.gl/en/docs/security/87fb8d47/): How to read binaries.
- [10. ECB Cut-and-Paste Attack Notes](https://ql.gl/en/docs/security/8b49314c/): Because AES-ECB encrypts each block independently, identical plaintext blocks map to identical
- [18. A Quick Guide to Zip Slip Vulnerabilities](https://ql.gl/en/docs/security/8bd9894c/): Zip Slip is a vulnerability that enables arbitrary file writes when ../ paths are not properly removed during archive extraction. If a web upload feature extracts ZIP files, an entry such as ../../../../var/www/html/shell.jsp can overwrite files on the server.
- [53. Exploiting CTF Binaries with Pwntools](https://ql.gl/en/docs/security/8f5ed561/): Binary exploitation is one of the most challenging yet rewarding CTF categories. Understanding the underlying vulnerability matters, but having the right tools can make the difference between solving a challenge and getting stuck. Enter pwntools, the Python library that has become the de facto standard for CTF binary exploitation.
- [06. Shadow DOM XSS Hunting Points](https://ql.gl/en/docs/security/9214b11b/): Modern frontends use Web Components to encapsulate UI inside the Shadow DOM. However, some sanitizers inspect only the light DOM. In that case, injecting unsafe HTML through a root exposed with attachShadow({mode:"open"}) can lead to XSS.
- [12. Accelerating Web Reverse Engineering with Source Maps](https://ql.gl/en/docs/security/9292b14f/): Even when a production JavaScript bundle is obfuscated, leaving a .map file behind exposes the original structure. From an attacker's perspective, it is a shortcut to finding hidden APIs, feature flags, and debug routes*.
- [04. A Quick Feel for SROP: Setting Every Register with sigreturn](https://ql.gl/en/docs/security/96aab283/): SROP (Sigreturn Oriented Programming) calls sysrtsigreturn to make the kernel restore the sigcontext on the stack verbatim. It can shorten a ROP chain or set all registers at once when gadgets are scarce. The key is to place a fake signal frame on the stack and trigger rt_sigreturn with syscall; ret.
- [34. Hash Length Extension: The Key to Bypassing MD5/SHA1 Signatures](https://ql.gl/en/docs/security/99f92404/): A web signature of the form hash(secret || msg) is vulnerable to length extension. Using the
- [63. XXE Injection: When XML Becomes an Attack Vector](https://ql.gl/en/docs/security/a33c9e00/): XML External Entity (XXE) Injection is an attack that abuses an XML parser's external-entity processing. It is a classic that has consistently appeared in the OWASP Top 10, yet it is still found in countless services.
- [47. Format String Vulnerabilities: A Recurring CTF Bug](https://ql.gl/en/docs/security/ab561a29/): Format string vulnerabilities are old bugs, but they remain a staple of CTFs. Once you understand the mechanism, they can enable everything from memory disclosure to arbitrary writes.
- [07. A Quick Feel for AES-CBC Bit-Flipping Attacks](https://ql.gl/en/docs/security/ac58c4f4/): In CBC mode, the preceding ciphertext block is XORed into the plaintext of the next block. This
- [08. Quickly Identifying Binary Protections with checksec in CTFs](https://ql.gl/en/docs/security/ae09dbea/): When solving reversing or pwn challenges, the task for the first 30 seconds is to inspect the protection options with checksec. Whether NX, PIE, and Canary are enabled completely changes the exploitation strategy.
- [32. CTF Crackme Reversing Patterns in 5 Minutes](https://ql.gl/en/docs/security/aeb73ed6/): Crackmes repeat patterns that can be solved with static analysis alone. Running through the checklist below in Ghidra/IDA is enough to speed up the process.
- [25. ret2dlresolve: A ROP Trick That Turns the Symbol Resolver Against Itself](https://ql.gl/en/docs/security/b5c23f4b/): ret2dlresolve abuses the dynamic linker's (ld.so) symbol-resolution process to call through ROP even functions that are absent from the binary. It is especially powerful in CTF binaries with limited PLT/GOT entries.
- [30. A Routine for Solving CTF Crypto Puzzles with Z3](https://ql.gl/en/docs/security/b5d2a0e2/): Cryptography CTFs often reach a point where "the equations are there, but the answer will not come
- [19. A Taste of Symbolic Execution with angr](https://ql.gl/en/docs/security/bbb26745/): In reverse engineering, symbolic execution is a technique that finds valid solutions by treating input conditions as variables and tracing execution paths. It is especially powerful for CTF crackmes and obfuscated branches.
- [58. SSTI: Complete Analysis of Server-Side Template Injection](https://ql.gl/en/docs/security/c7a5d144/): SSTI (Server-Side Template Injection) is a vulnerability that occurs when a template engine interprets user input as code. It looks like XSS, but it executes on the server and can lead to RCE.
- [03. A Fast Browser Extension Reversing Routine](https://ql.gl/en/docs/security/c9164bba/): For Chrome/Edge browser extensions, extracting CRX → ZIP gets you almost to the finish line. Even when the source is obfuscated, establishing the structure first dramatically speeds up analysis.
- [65. Mastering SQL Injection — An Essential Web Hacking Skill](https://ql.gl/en/docs/security/c98b1c55/): TL;DR: SQL Injection remains one of the most dangerous web vulnerabilities. This practical guide
- [24. The Core Flow of ELF PLT/GOT Hijacking](https://ql.gl/en/docs/security/d2f35e71/): In reversing and CTFs, a GOT overwrite remains a powerful bypass point. Once the dynamic-linking flow is clear, you can cleanly redirect the call path.
- [62. XSS Hunting: From Basics to Bug Bounties](https://ql.gl/en/docs/security/dc777f86/): Cross-Site Scripting (XSS) remains one of the most common vulnerabilities in web applications today. Although it has been known for decades, developers keep repeating the same mistakes, making XSS hunting a profitable skill for bug bounty hunters and CTF players alike.
- [31. Binary Diffing: A Fast Routine for Finding Vulnerabilities in Patches](https://ql.gl/en/docs/security/dca51669/): When a patch is released, the fastest vulnerability-analysis routine is binary diffing. Knowing only “which function changed” can reveal the attack point.
- [55. ROP Chains: How to Bypass a Stack Canary](https://ql.gl/en/docs/security/e267f8e3/): Modern binaries are protected by NX (Non-Executable Stack) and Stack Canary.
- [60. Essential Web Security Testing Tools](https://ql.gl/en/docs/security/e296f6ad/): Knowing good tools is fundamental to web hacking. Whether for CTFs or real bug bounties, these tools are essential.
- [50. Heap Exploitation: An Introduction to tcache Attack Techniques](https://ql.gl/en/docs/security/e325cb15/): TL;DR: Understanding glibc's tcache gets you halfway through a heap exploit. This article summarizes
- [05. ret2csu Mini Cheatsheet](https://ql.gl/en/docs/security/e3e3204d/): When ROP gadgets are scarce in a PIE/NX environment, you can build a function call with two gadgets
- [43. CORS Misconfigurations and Practical Exploitation](https://ql.gl/en/docs/security/ee2f2b16/): CORS (Cross-Origin Resource Sharing) is a browser mechanism that controls access to resources from other origins. A misconfiguration can let an attacker steal an authenticated user's data.
- [33. CSP Bypass Notes: nonces, strict-dynamic, and Mistakes](https://ql.gl/en/docs/security/ef85eb3f/): CSP reduces XSS, but configuration mistakes immediately become bypass opportunities. These are only
- [67. Function Hooking: A Reverse Engineer's Secret Weapon](https://ql.gl/en/docs/security/f2cced46/): Function hooking may sound intimidating at first, but it becomes an incredibly powerful tool once you understand it. From malware debugging and anti-cheat bypasses to CTF challenges, hooking lets you intercept and modify a program's behavior at runtime.
- [15. Finding Reverse-Engineering Clues in Ten Minutes with strace and ltrace](https://ql.gl/en/docs/security/fcc266ba/): Running strace and ltrace before static analysis gives you an immediate view of I/O flow and library calls. They are excellent for narrowing down key-verification routines quickly in CTF crackmes.
- [20. Notes on Bypassing ptrace Anti-Debugging](https://ql.gl/en/docs/security/ff69338b/): A reverse-engineering target will often terminate immediately when it detects a debugger with ptrace(PTRACE_TRACEME). The key is to control the call site and return value.

## Docs: tools

- [00. Tools](https://ql.gl/en/docs/tools/): // TOOLING — Development tools, editors, terminals
- [Neovim Keymap Inventory and Troubleshooting](https://ql.gl/en/docs/tools/0dd0d73d/): This document expands on a personal note from 2023—a shortcut table I wrote so I would not forget it.
- [ITERM2 shell integration script](https://ql.gl/en/docs/tools/27df7163/): P4R4D0XB0X — Systems Engineer, Rust Developer, AI/ML Engineer. Low-level systems, infrastructure, and AI.
- [FNM Fast Node Manager](https://ql.gl/en/docs/tools/8e9d0608/): Purpose
- [Deep Dive Git Cli](https://ql.gl/en/docs/tools/9a881b9b/): Written with macOS in mind.
- [When Korean Text Is Unreadable](https://ql.gl/en/docs/tools/fa5c8f52/): P4R4D0XB0X — Systems Engineer, Rust Developer, AI/ML Engineer. Low-level systems, infrastructure, and AI.

## Docs: typescript

- [00. Typescript](https://ql.gl/en/docs/typescript/): // FULLSTACK — Web development on top of a type system
- [01. Grammar](https://ql.gl/en/docs/typescript/94bae768/): TypeScript Cheatsheet

## Docs: writing

- [00. Writing](https://ql.gl/en/docs/writing/): // EDITORIAL-SYSTEM — Write professional technical prose, but reinforce hard terminology with plain-language explanations and examples.
- [01. Blog Editorial Guide](https://ql.gl/en/docs/writing/5668888c/): Monitoring sources, writing prompts, term explainers, and image and attribution standards for a professional technical blog.

## Blog

- [AgentsCAD: Automated Design for Manufacturing of FDM Parts — Multi-Agent LLM Reasoning and Geometric Feature Recognition](https://ql.gl/en/blog/01427437/): A technical review of AgentsCAD: research that automates design-for-manufacturing (DFAM) modifications for FDM by combining STEP B-Rep parsing, overhang detection, GraphSAGE-based semantic label injection, multi-agent LLM reasoning with Claude Sonnet, and GPT-4o visual verification. It also states the uncertainty where evidence is limited.
- [Mathematics for Computer Science (MCS) — Structure and Core Concepts](https://ql.gl/en/blog/0667fd2f/): An analysis of the purpose, core concepts such as proof methods, axioms, and induction, organization, and logical progression of MIT's Mathematics for Computer Science course notes (Lehman, Leighton, Meyer, 2018). It summarizes the table of contents and excerpts in the evidence pack while stating uncertainties.
- [Mamba-3: Connecting Linear Inference to Practical Sequence-Modeling Quality](https://ql.gl/en/blog/14e25d3e/): Mamba-3 seeks to improve both efficiency and state-tracking ability from a state-space perspective through better recurrence, complex-valued state updates, and multi-input/multi-output I/O. This article analyzes the design and evidence boundary, focusing on the 1.5B results stated in the paper's abstract.
- [Mamba: Approaching Content-Based Reasoning with Selective State Spaces](https://ql.gl/en/blog/1d139737/): This article explains why Mamba targets both long-sequence efficiency and language-modeling performance through selective SSMs that change state-space parameters based on the input and hardware-aware parallel algorithms.
- [Resona: Using Retrieval to Improve Context Copying in Linear Recurrent Models](https://ql.gl/en/blog/23360387/): Resona is a framework that retrieves needed information from the input context and reinjects it into a linear recurrent language model. Based on the paper's abstract, this article examines the bottleneck in context copying, the meaning of combining retrieval with recurrence, and the evaluation, cost, and evidence that must be checked before practical adoption.
- [RWKV-5 Eagle and RWKV-6 Finch: Expanding the State into a Matrix and Making Recurrence Dynamic](https://ql.gl/en/blog/2a987eb6/): This article explains, with evidence from the original paper, how Eagle and Finch retain RWKV-4's efficient inference path while adding matrix states, multiple heads, and data-dependent recurrence.
- [Five Recent Signals Across the AI Stack: Agents, Voice LLMs, EmbeddingGemma, Local sLLMs, and Qiskit Paulice](https://ql.gl/en/blog/2ff9ec4f/): One recently verifiable public source from each of AI, LLMs, embeddings, sLLMs, and quantum computing, with a summary of its technical significance and limitations.
- [Claw Patrol: Design and Operational Perspectives on a Firewall for Agents](https://ql.gl/en/blog/3035ac8f/): A README-based review of Claw Patrol's architecture, HCL and CEL rule representation, deployment modes, and operational considerations for protecting agents such as LLMs and automated processes.
- [Valmis: Analysis of a Security-Centered AI Agent Platform](https://ql.gl/en/blog/33a0a1e1/): An analysis of Valmis as a security-centered AI agent platform based on its GitHub repository. It provides a technical overview of core designs and operational considerations including proxy-based credential protection, container isolation, multistep workflows, and browser automation.
- [The Legendre Transform: Intuition, Examples, and the Thermodynamic Connection](https://ql.gl/en/blog/3551c399/): A detailed explanation of the Legendre transform's intuitive meaning and mathematical definition, how it works through a harmonic-potential example, and its connection to partition functions and free energy through the Laplace transform. Each core idea follows a step-by-step problem–intuition–meaning progression.
- [Hidden State Poisoning: When Mamba's Efficient State Becomes an Attack Surface](https://ql.gl/en/blog/390c6aee/): Hidden State Poisoning Attacks against Mamba-based Language Models analyzes how a short trigger can overwrite the hidden state of Mamba-family models and break information retrieval. Centered on HiSPA and RoBench-25, this article explains the relationship between efficient state and the security boundary within the scope of the original abstract.
- [S4: How to Read Long Sequences with Structured State Spaces](https://ql.gl/en/blog/4a6889f1/): This article examines how the Structured State Space sequence model (S4) structures state space models to handle long dependencies and reduces them to computable Cauchy kernel operations. It distinguishes only the claims confirmed by the arXiv abstract and metadata.
- [uv: An Analysis of the Rust-Built, Extremely Fast Python Package and Project Manager](https://ql.gl/en/blog/4aeb8d0a/): Astral's uv is a Python package and project manager written in Rust. Working from the README and repository evidence, this post analyzes its core design, its performance claim (10–100x), its features (project/script/tool/Python-version management, a pip-compatible interface), and the implementation trade-offs. Verify the detailed behavior against the official documentation.
- [xLSTM: How to Scale Recurrent Memory Again](https://ql.gl/en/blog/4d111147/): xLSTM redesigns LSTM gating and memory structures from the perspective of scaling modern language models. Based on the public abstract and metadata, this article examines the roles of sLSTM and mLSTM, the meaning of parallelization and memory state, and performance conditions that remain unverified.
- [AISBF — Operating an OpenAI-Compatible Router and Local CoderAI Workers](https://ql.gl/en/blog/4e5a88ed/): A review of the practical implications of AISBF's OpenAI-compatible gateway routing, local CoderAI worker integration, and privacy and failover policies. It analyzes demo and self-hosting options and operational considerations from the available technical evidence.
- [EAGLE-360: Embodied Active Global-to-Local Exploration in 360° Environments](https://ql.gl/en/blog/53d04389/): The 2026 EAGLE-360 paper proposes a Global-to-Local strategy, RoPE Rolling positional encoding, and an SFT plus GRPO training pipeline for active exploration in 360° panoramic spaces. Based on the public abstract, this post provides a technical overview of the contributions and design, explaining both the evidence and uncertainty around the stated dataset and performance claims.
- [prompts.chat Analysis: An Open-Source Prompt Library with Self-Hosting Options](https://ql.gl/en/blog/55c80852/): An analysis of features, licenses, self-hosting, and integrations based on the prompts.chat GitHub repository and related site. It organizes core facts and uncertainties within the supplied evidence.
- [Mira — A Self-Hosted AI Code Reviewer: Overview, Deployment, and Operations](https://ql.gl/en/blog/6c19ded8/): A technical overview of the self-hosted code-review pipeline, indexing and vulnerability scanning, LLM integration, and operational considerations based on Mira's Apache-2.0 open-source repository. It reviews repository-backed claims about features, deployment, and benchmarks and identifies unclear areas.
- [Automated Grading of Linux/Bash Exams with LLMs — Evaluating a Four-Level Cognitive Taxonomy](https://ql.gl/en/blog/6edda312/): A research summary and practical analysis of automated grading for short Linux/bash answers using large language models. It focuses on how a four-level cognitive taxonomy and rubric-based prompts affect results.
- [Open WebUI: Technical and Product Analysis of a Local AI Interface and Self-Hosted Platform](https://ql.gl/en/blog/70b7fd97/): A technical analysis of Open WebUI's architecture, deployment, major features, and operational considerations such as RAG, vector databases, authentication, and observability, based on a summary of its GitHub repository. Details absent from the documentation are explicitly marked as uncertain.
- [Qdrant: Overview of a High-Performance Vector Search Engine](https://ql.gl/en/blog/76603fb9/): An overview of Qdrant's architecture and core features, including vector and hybrid search, edge deployment, and quantization, along with key technical considerations for implementation and operation. The source is the qdrant/qdrant GitHub repository.
- [trajeckt: A Sealed-Commitment Runtime Enforcement Gateway for Agents](https://ql.gl/en/blog/7694284a/): An analysis of the trajeckt repository: its runtime gateway design for blocking multistep trajectory-style agent exploits, core mechanisms including sealed graphs, global trajectory checks, and off-agent state, plus deployment and integration techniques and limitations.
- [Why Does Mamba-2 Resemble the Transformer? Reading SSMs Through SSD](https://ql.gl/en/blog/76b0dce0/): Following Tri Dao and Albert Gu's Mamba-2 paper, this article explains the connection between state space models and attention, the hardware implications of the SSD algorithm, and the claims that remain unverified.
- [Hermes Agent: Analyzing a Self-Learning AI Agent Framework for Practical Deployment](https://ql.gl/en/blog/77fddbfc/): A technical overview of the architecture, operating options, installation flow, and security and operational caveats visible in the Nous Research Hermes Agent repository. It analyzes the core design and deployment choices from public documentation and README evidence.
- [Program-as-Weights: A Programming Paradigm for Fuzzy Functions](https://ql.gl/en/blog/7d886a91/): The main idea and technical implications of the 2026 arXiv paper Program-as-Weights. It reviews PAW's approach to compiling fuzzy functions defined in natural language into small, reproducible neural artifacts that run locally, efficient execution through an ultralight interpreter, and the release of FuzzyBench.
- [H3: The Memory and Comparison Capabilities SSMs Missed in Language Modeling](https://ql.gl/en/blog/7ecdd9d9/): Hungry Hungry Hippos decomposes the linguistic weaknesses of state space models into token recall and comparison, then explains how H3, FlashConv, and hybrid attention combinations address them.
- [Adoption and Ecosystem Health of Open-Source Multi-Agent Frameworks: A Longitudinal Analysis](https://ql.gl/en/blog/82fb30d2/): Based on the key findings of the arXiv paper 'Adoption and Ecosystem Health,' this article compares and interprets popularity metrics for open-source multi-agent (agentic) frameworks against actual ecosystem health, including contributor density, cross-project contribution, and retention. It offers a practical perspective so engineers and product teams do not rely on star counts alone when selecting a framework.
- [The First Question After the Transformer: Are RNNs Really Finished?](https://ql.gl/en/blog/906ca0d1/): Using an ICLR discussion of Transformer alternatives as an example, this article summarizes the core claims and limitations of recurrent models that have regained attention since the 2017 Transformer.
- [The NautilusTrader Engine: Analyzing Rust-Native Research-to-Execution Parity](https://ql.gl/en/blog/90d159a5/): An analysis of open-source NautilusTrader's structure and design philosophy based on its README. It covers the roles of the Rust core and Python control plane, its event-driven deterministic simulation model, and its adapter integration strategy.
- [How Reasoning Effort Affects First-Try Reliability in Agentic Code Generation More Than Tool Access — An Observational-Study Summary](https://ql.gl/en/blog/993443a1/): A summary and analysis of 90 observed agent runs and their practical implications, based on Achint Mehta's 2026 arXiv paper Reasoning effort, not tool access, buys first-try reliability in agentic code generation. The central result reports that reasoning effort and model capability tier decisively affect first-try completeness, while external testing tools increased cost without improving reliability.
- [CheatSheet as Operational Memory](https://ql.gl/en/blog/9c87d87f/): A way to turn frequently searched commands and decision criteria into personal operational memory. The article stays professional while explaining key terms with simple examples.
- [Is Recurrent State a New Attack Surface?](https://ql.gl/en/blog/9fb6baf2/): This article treats recurrent state and agent memory as security boundaries. It connects Mamba, Hidden State Poisoning, trajeckt, and Valmis while separating paper evidence from product and security context to examine the risks of state poisoning, reuse, and data flows.
- [Textbooks Are All You Need — Improving Small Code LLMs with High-Quality Textbook Data](https://ql.gl/en/blog/a3babb9c/): A summary and analysis of Microsoft Research's Textbooks Are All You Need. It examines how the 1.3B-parameter code LLM phi-1 substantially improved HumanEval and MBPP performance through textbook-quality data and a small synthetic exercise dataset, along with the method's limitations.
- [Does SSM Replace the Transformer, or Do They Divide the Work?](https://ql.gl/en/blog/a7c32433/): This article reads the competition between state space models and Transformers as a question of division of labor rather than structural replacement. Based on the abstracts and publicly stated claims of Mamba, Griffin, Jamba, and Mamba-3, it organizes the trade-offs among long contexts, generation, and hardware efficiency.
- [Is a Fixed-Size Hidden State Memory, or Compression?](https://ql.gl/en/blog/ac4b2e1e/): This article distinguishes fixed-size recurrent hidden states as memory devices versus information compression. It explains the role of state through S4, Mamba, RWKV, and Resona, while treating Qdrant and Statey separately as external-memory contexts.
- [Overview of the LangChain Agent Engineering Stack](https://ql.gl/en/blog/ad9791a2/): A technical overview of the core concepts and ecosystem for building agent-based LLM applications, based on LangChain's README. It focuses on components, strengths and weaknesses, and practical adoption considerations.
- [Statey: A Shared Database Inside AI Chat, Based on MCP](https://ql.gl/en/blog/c5fd0099/): A technical analysis of Statey's design as a record-based database centered on AI chat, including integration implications, limitations, and evaluation points.
- [Firecrawl: A Web-Scale Data Collection Platform from an Infrastructure Perspective](https://ql.gl/en/blog/cd78c9d9/): A summary based on the firecrawl/firecrawl GitHub repository documentation, covering its role as a web-context API, infrastructure considerations such as proxies, orchestration, and latency, and agent and SDK features from technical and product perspectives.
- [When LLM Personas Do and Do Not Support Visualization Design — A Cross-Model Study of Color Assignment and Chart Choice](https://ql.gl/en/blog/cf5858f1/): An analysis of a 2026 arXiv paper on how LLM personas affect visualization color assignment and chart-idiom selection, including model-specific differences and practical recommendations.
- [Forge: Analyzing the Tool-Calling Reliability Layer for Self-Hosted LLMs](https://ql.gl/en/blog/d14f9aa0/): A technical overview of Forge's design and core guardrails for improving tool-calling reliability in self-hosted LLM environments, based on the antoinezambelli/forge repository README and documentation. It summarizes production-adoption considerations and implementation concerns.
- [HiPPO: Recurrent Memory with Optimal Polynomial Projections — A Paper Commentary](https://ql.gl/en/blog/d3b70ba2/): A step-by-step explanation of how the HiPPO framework reformulates recurrent neural-network memory as online function approximation, including the intuition, mathematics, and experimental advantages of the LegS (Scaled Legendre) update.
- [A Paper Roadmap for Understanding Surrogate Gradients](https://ql.gl/en/blog/d657d100/): A reading order covering SNN modeling, backpropagation, STDP, SpikeProp, ANN-to-SNN conversion, SuperSpike, SLAYER, and STBP to understand surrogate gradients, which train the discontinuous spikes of spiking neural networks through backpropagation.
- [ReContext: Improving Long-Context Reasoning through Recursive Evidence Replay](https://ql.gl/en/blog/d7d549fb/): A summary and analysis of RECONTEXT, a training-free inference technique for helping LLMs use relevant evidence in long contexts more effectively. It describes the method's core, theoretical perspective through associative memory, experimental overview, limitations, and applicability.
- [AutoGPT Platform Analysis: Agent Platform Architecture and a Practical Self-Hosting Guide](https://ql.gl/en/blog/d9f416e3/): A technical overview of the AutoGPT platform's autogpt_platform components, self-hosting flow, major tools including Forge, agbenchmark, the frontend, and CLI, and licensing considerations, based on the Significant-Gravitas/AutoGPT README and documentation excerpts. Because some source details are partial, consult the official documentation as well.
- [Griffin's Compromise: Combining Gated Linear Recurrence with Local Attention](https://ql.gl/en/blog/dd8d73cd/): This article separates the evidence reported by the paper from the open questions of adoption, explaining how Hawk and Griffin combine recurrent-model efficiency with the Transformer's representational power.
- [Neuron-Aware Data Selection for Annotation-Free LLM Self-Distillation — Summary and Technical Commentary](https://ql.gl/en/blog/e884d0c3/): A commentary based on the abstract of the arXiv paper 'Neuron-Aware Data Selection for Annotation-Free LLM Self-Distillation.' It covers the concepts behind neuron-activation-based data selection and on-policy self-distillation teacher construction, the claimed effects, and remaining uncertainties.
- [DemoPSD: An Analysis of Disagreement-Modulated Policy Self-Distillation](https://ql.gl/en/blog/eabec12f/): A summary of DemoPSD's core idea and implications: selective adoption of teacher guidance based on disagreement, a reverse-KL barycenter target, and theoretical claims and experimental results concerning privileged-information leakage and preservation of exploration. It is grounded in arXiv 2607.02502v1 and marks details outside the supplied evidence as uncertain.
- [Best AI Agent Red Teaming Tools in 2026: Features, Limitations, and Adoption Considerations](https://ql.gl/en/blog/edbc94f6/): A comparison and evaluation framework for agentic AI red-teaming tools in 2026. It focuses on integrated security and quality, agent-native testing, vulnerability-response pipelines, and organizational process, while stating the scope and limitations of the source material.
- [Cloudflare workerd Runtime Analysis: A Guide to the Server-Side JavaScript/Wasm Environment](https://ql.gl/en/blog/f0d9a686/): An overview of the design principles, use cases, build, configuration, and deployment flow, and security considerations of Cloudflare's open-source server-side JavaScript/Wasm runtime workerd, based on the cloudflare/workerd GitHub repository.
- [Will Scaling Improve Social Simulation with LLMs?](https://ql.gl/en/blog/fb92b6be/): An explanation of the key points and practical implications of the arXiv paper 'Will Scaling Improve Social Simulation with LLMs?' It focuses on scaling-law trends across social-simulation subdomains including opinion modeling, behavioral simulation, and longitudinal forecasting, then discusses limitations and future research directions.
- [Mamba-3: Improved Sequence Modeling Using State Space Principles — Key Ideas and Technical Implications](https://ql.gl/en/blog/fcd5a843/): A summary of the Mamba-3 paper based on an evidence pack, explaining its three core SSM improvements—exponential-trapezoidal discretization, complex-valued state updates, and MIMO—in technical context. It focuses on the implications of its inference-first design, hardware efficiency, and verifiable results.
- [Gemini CLI: Analyzing Terminal Agent Workflows and Integration Strategies](https://ql.gl/en/blog/fe58bfd2/): A technical analysis of the terminal-based agent workflows, authentication options, MCP integrations, and automation use cases provided by Google's open-source Gemini CLI, based on the official GitHub repository README and documentation.