All projects

A from-scratch LSM-tree in Go: skip-list memtable, Bloom-filtered mmap SSTables, three live-switchable compaction strategies, and Raft quorum replication with multi-shard rebalancing behind one API.

A B+Tree engine in Go with MVCC snapshot isolation, ARIES-style WAL crash recovery, slotted 4KB pages, a Clock-eviction buffer pool, and group commit. The low-write-amplification counterpart to the LSM engine.

A complete SQL engine built end to end: lexer, recursive-descent parser, analyzer, a rule-based optimizer, cost-based join ordering by dynamic programming, and a Volcano-model executor with an interactive plan visualizer.

A transaction engine running 2PL and MVCC side by side: serializable snapshot isolation, ARIES recovery, wait-for-graph deadlock detection, and Raft replication, with a browser playground for reproducing isolation anomalies.

A replicated key/value store on a from-scratch Raft: linearizable reads via ReadIndex (no log write), streaming snapshots, joint-consensus membership changes, and correctness checked with Porcupine linearizability testing.

An interactive causality lab: Lamport, vector, and matrix clocks, dotted version vectors, Chandy-Lamport global snapshots, and BSS causal delivery, all rendered as live space-time diagrams with eight failure scenarios.

Single-leader, multi-leader, and leaderless (Dynamo-style) replication in one simulator: tunable N/W/R quorums, read repair, hinted handoff, Merkle anti-entropy, and a linearizability checker over the resulting histories.

An observability backend ingesting OTLP and native spans with seven pluggable samplers (including tail-based and adaptive) and RED metrics computed over sliding-window histograms, streamed to a React waterfall UI.

An HTTP reverse proxy and load balancer in pure stdlib Go: eight balancing algorithms (including peak-EWMA and bounded-load consistent hashing), circuit breaking, request hedging, SNI multi-cert TLS, and live config reload.

An Airflow-semantics workflow engine in pure Go, no Python or broker: YAML DAGs, 11 trigger rules, poke/reschedule sensors, cycle detection, hot reload via fsnotify, a dead-letter queue, and a React Flow UI.

A self-hosted feature-flag platform: microsecond local evaluation, real-time SSE flag updates, statistically-sound A/B testing (mSPRT), and Go plus TypeScript SDKs kept in lockstep by 45+ cross-SDK conformance fixtures.

A resilience toolkit with six rate-limiting algorithms (token bucket, GCRA, sliding window, leaky bucket, fixed window, adaptive), plus circuit breaker, bulkhead, and retry budget. Zero-dependency core, optional Redis backend.

A userspace M:N fiber scheduler with work-stealing across per-worker run queues, fiber-aware mutexes, channels, and wait groups, priority scheduling with anti-starvation aging, and a live scheduler dashboard.

Lock-free data structures built directly on CAS: a Treiber stack, Michael-Scott queue, Harris-Michael sorted list, and cache-line-padded SPSC/MPMC ring buffers, with linearizability verified by model-checked tests.

An inode-based filesystem (ext2-like) with ext3-style ordered journaling and crash recovery, a write-back buffer cache with CRC32 commit records, and a real-time SVG block-map visualizer plus a 20-command shell.

A multi-provider agent orchestrator with autonomous loops, cron, parallel swarms, hard token budgets, and a crash-safe SQLite state store. Recently ported from a Python control plane to a native Go runtime. On PyPI + npm.

An LLM-evaluation and prompt-optimization pipeline: a calibrated LLM-as-judge, a statistical deploy gate (paired-bootstrap CI with Holm correction), and five optimizers including a from-scratch GEPA. On PyPI as promptline-opt.

A full-stack resume optimizer that rewrites LaTeX against a target job description, streams the compile, and scores ATS keyword coverage with pgvector semantic matching, showing exactly what a parser reads back.

A retrieval-augmented knowledge base: a FastAPI backend ingests PDFs, DOCX, and images into Pinecone, and GPT-4 answers questions with citations back to the source chunks. React search and chat UI on top.

A multi-tenant payment platform in Go using double-entry ledgering, a transactional outbox, and saga orchestration with explicit per-entity state machines, three-way reconciliation, and dead-letter handling.

A cross-platform enterprise backup product: Go backend, Next.js web, Rust/Tauri desktop, React Native mobile, and browser extensions, with Ansible and Helm deployment and OPA policy-as-code.

A high-performance P2P file-sharing CLI in C++17: an epoll event loop handling 200+ concurrent peers, a custom binary protocol with CRC32 validation, resumable transfers, and a worker thread pool.

A geospatial analysis chatbot on Next.js wiring Google Earth Engine, Vertex AI, and Supabase: RAG over map data, live MapLibre rendering, vector uploads, land-cover mapping, and change detection.

A full-stack AI ad generator: scrape a product URL, generate a GPT-driven script, and render a 9:16 vertical video with FFmpeg. FastAPI backend with a React and Vite front end.

A real-time malware scanner pairing a Chrome extension with a Flask backend and the VirusTotal API, with an encrypted quarantine store, a management dashboard, and email alerts.

A pure-Python linear algebra library with no NumPy underneath: vectors, matrices, and the decompositions that matter for ML (QR, SVD, Cholesky, eigendecomposition, PCA) plus Gram-Schmidt projection. Phase 0 of a from-scratch ML stack.

Automatic differentiation, symbolic differentiation, numerical integration, and Taylor expansion in pure Python. A small autograd built to understand exactly what a real one is doing.

Probability and statistics from scratch in pure Python: distributions, MCMC samplers, Bayesian estimation, and information theory, with no SciPy underneath. Part of the same from-scratch ML foundation.

Gradient-based optimizers, learning-rate schedules, line search, and second-order and constrained methods, implemented in pure Python to watch how each one actually converges.

A five-milestone deep-learning project for music genre classification and mashup detection: audio feature pipelines, CNN and RNN models, evaluation, and a Flask deployment.