< Projects />

PUBLISHED COMPLETED

AI/ML

The Recall Ceiling — Why LLM Recommendation Reranking Fails in Practice

Retrieval recall, not reranking sophistication, sets the ceiling on two-stage recommendation. Proved as a combinatorial…

Python PyTorch LightGCN GraphSAGE +2

EXPLORE()

PRIVATE COMPLETED

AI/ML

Triton-Kernel-Optimization

Production-grade Grouped Query Attention (GQA) in Triton with composable attention patterns: 8 variants via tl.constexp…

Python Triton 2.1+ PyTorch 2.0+ CUDA +1

EXPLORE()

COMPLETED

Blockchain

SimpleBTC

A complete Bitcoin implementation in ~3000 lines of Rust — full UTXO model, secp256k1 ECDSA, M-of-N multisig, BIP125 RB…

Rust secp256k1 RocksDB Rayon +1

EXPLORE()

PRIVATE COMPLETED

Systems

Infer-GC

GC-aware LLM inference engine applying garbage collection semantics (reference counting, generational collection, memor…

Python PyTorch CUDA PagedAttention +1

EXPLORE()

COMPLETED

Systems

Vulkan-Study

Interactive Vulkan learning system with 20 progressive exercises and a final renderer project

C++17 Vulkan 1.2+ GLFW GLM +2

EXPLORE()

LIVE DEMO COMPLETED

Tools

Foobar2026-WASM

Browser-based music style filter and audio processing workstation with real-time effects

Rust WebAssembly React TypeScript +2

EXPLORE()

COMPLETED

AI/ML

MemoryForge

Three-layer cognitive memory for LLM agents -- working, episodic, and semantic -- with 473 passing tests, hybrid retrie…

Python FastAPI Qdrant Neo4j +5

EXPLORE()

PRIVATE COMPLETED

Systems

s3-like-storage

An S3-compatible distributed object store with a purpose-built LLM KV-cache engine: radix tree keyed on token prefixes,…

Go 1.21+ PostgreSQL Redis Nginx +2

EXPLORE()

COMPLETED

Systems

Timeline-Feed — Fan-Out-on-Write, and an Audit of Its Own Gaps

A fan-out-on-write timeline service with the read path, the write path and the failure modes separated — and an analysi…

TypeScript Node.js MongoDB Redis +3

EXPLORE()