All notes
39 entriesAI Notes2 notes
Vibe-Coded, Not Vibe-Validated
How I use AI as an execution amplifier while keeping premises, constraints, validation, and conclusions human-owned.
AI Wrote Everything Except This
A mostly raw field note on using Codex to revive half-finished projects, write blog drafts, optimize code, fight GUI layouts, and avoid doing every chore by hand.
Project Lessons16 notes
ClusterLens Algorithms: Clustering Image Embeddings From First Principles
A beginner-first, experimental tutorial for PCA, KMeans, MiniBatchKMeans, FAISS, HDBSCAN, similarity graphs, outliers, reranking, and cluster quality.
From Pixels and Words to Semantic Search
A beginner-first tutorial on representations, pixels, features, embeddings, image and text encoders, shared vector spaces, and measured CLIP, SigLIP, and DINO experiments.
Perceptual Hashing And Duplicate Review
A from-scratch visual tutorial on exact hashes, dHash, pHash, wHash, thresholds, indexing, embedding reranking, ORB, and safe duplicate review.
Face Clustering From Pixels To People
A from-scratch visual tutorial on face detection, landmarks, alignment, embeddings, thresholds, clustering, prototypes, and human review, backed by an actual ClusterLens run.
ClusterLens UI Runtime: Virtualized PyQt, Workers, And Generated State
A desktop runtime note on making local ML image tools responsive with PyQt virtualization, lazy thumbnails, worker jobs, generated storage, and safe file actions.
Neural Network Tutor: From Replay Lab to Production Debugger
A story-driven visual course that lets every neural-network idea earn its place by solving a visible problem, from one weighted vote through hidden representations, transformers, modern learning phenomena, and production debugging.
Email Before AI: MIME, Canonical Text, Multilingual Processing, and Privacy
A beginner-first tutorial following email from MIME bytes through canonical text, language detection, translation, Unicode offsets, PII detection, redaction, and safe model input.
Email Classification and Personal AI: Rules, Naive Bayes, Transformers, Specialists, and Evaluation
A from-zero visual tutorial building email classification from rules and Naive Bayes through personalization, specialist-model cascades, confidence, local model evaluation, and safe promotion.
From Repeated Email to Structured Knowledge: Templates, Extraction, Facts, and Summaries
A from-scratch tutorial following repeated email through template mining, source-linked extraction, symbolic facts, proof trees, and hierarchical summarization.
RAG From Raw Email To Grounded Answers
A from-scratch visual tutorial following one email through tokenization, embeddings, exact and hybrid retrieval, evidence packing, grounded generation, GraphRAG, security, and evaluation.
Inside ThunderbirdAI: From Raw Email to a Grounded AI Answer
An implementation-level walkthrough of how ThunderbirdAI queues mail, builds canonical local records, indexes evidence, enforces scope, and uses an optional endpoint without treating the model as the mailbox.
Desktop Local App Lessons
A desktop product note on workers, virtualized PyQt tables, import auditability, parser health, schedulers, dashboards, tray tools, and generated data.
Where a Website Meets Chromium
A process-by-process map of Chromium, the hardened fork's hook points, and how website-visible decisions cross Blink, Content, Chrome, services, and local automation.
Changing What a Website Can Observe
How Hardened Chromium substitutes visibility, focus, lifecycle, idle, screen, fullscreen, API, and cursor signals—and which secondary signals remain.
Protecting User Actions and Physical Devices
How the fork protects selection, context menus, keyboard defaults, clipboard, drag data, camera, microphone, location, and display capture.
Website View, Automation Signals, and the Shared Browser
How Website View policy, WebDriver signaling, launch-time switches, and an authenticated broker connect local applications to one visible Chromium instance.
Performance Notes9 notes
inih Optimization: Simpler Scans and Flat Storage
A benchmark-backed note on speeding up inih with bulk string reads, explicit ASCII scan classification, selective AVX2, and flat C++ reader storage.
zlob Optimization: Faster In-Memory Glob Matching Without More Layers
A benchmark-backed note on widening zlob's suffix fast path, pruning recursive work earlier, reusing outputs, adding count-only APIs, and threading only when the scan is big enough.
miniz Optimization: Checksums Win Before the Codec Does
A benchmark-backed note on CRC-32, Adler-32, SIMD, prefetching, memcpy, and why codec-level compression gains are harder than checksum wins.
Clean Code Tax, Measured
A benchmark-backed note on when clean-code idioms are free, when compilers erase them, and when abstraction becomes hot-path cost.
ClamAV SIMD Fast Paths For Parser-Heavy Scanning
A microbenchmark-backed note on accelerating ClamAV-style parsing work with SIMD string search, whitespace skipping, hash comparison, hardware CRC32, and base64 decoding.
Speeding Up pdftotext When You Only Want The Text
A benchmark-backed note on making PDF text extraction faster with cheaper utility output, lighter ordered text ingestion, image-skip logic, page threading, and an unordered direct-text mode.
Cache-SIMD-DSA
A standalone benchmark note for 10 DSA problems, with naive, cache-aware, SIMD, and cache-aware SIMD C++ variants measured from 4 KiB to 64 MiB.
ORB Python-To-Rust Speedup Record
A benchmark-backed record of Python-to-Rust ORB backtester speedups, accepted stage ports, rejected experiments, and claim boundaries.
Cache Lines Before Big-O: Optimizing Search Algorithms by Changing Memory Access
A single-system benchmark study of arrays, pointer chasing, binary search layouts, hash maps, prefetching, memory-level parallelism, and false sharing.
Computer Internals Notes7 notes
Linux I/O Path: From write() to NVMe Doorbell
A structure-by-structure map of Linux buffered I/O, direct I/O, VFS, page cache, iomap, bio, blk-mq, NVMe commands, DMA mappings, completions, and the PMEM/DAX contrast.
Mutex Internals: From Cache Lines to Futex Wait Queues
A hardware-to-runtime map of atomics, cache coherence, memory ordering, spinlocks, futex-backed mutexes, kernel mutex paths, and condition variables.
Packet Flow: NIC, DMA, Rings, Kernel, XDP, and DPDK
A hardware-to-userspace walkthrough of how packets move through NIC DMA rings, Linux drivers, NAPI, sk_buff, TCP, XDP, AF_XDP, and DPDK.
What Does Clearing .bss Mean in Early Kernel Boot?
A practical mental model for .bss: how compilers, linkers, ELF loaders, and early kernel boot cooperate to make zero-initialized globals real.
Booting, My Confusions and ChatGPT
A capability-by-capability note on early boot: reset vectors, firmware, microcode, XIP, cache-as-RAM, DRAM bring-up, C runtime setup, memory types, and first page tables.
Hardware Memory Path: From Virtual Address to DRAM, MMIO, and DMA
An interactive map of what can happen after a user instruction touches memory: translation, TLBs, caches, page walks, DRAM, MMIO, PCIe, IOMMU, and DMA.
SIMD Mapping: From Normal Code to Vector Code
A practical tutorial and cheatsheet for translating scalar loops, branches, reductions, and byte scans into AVX2 SIMD patterns.
Interview Notes3 notes
C Interview Notes
Deep-answer interview notes for C questions: memory, allocation, linkage, arrays, alignment, and synchronization.
C++ Interview Notes
Deep-answer interview notes for C++ questions: RAII, object lifetime, templates, casting, ownership, and memory order.
System Design Interview Notes
Deep-answer interview notes for system design questions: APIs, caches, Kafka, consistency, disaster recovery, and Spanner.