AI/ML engineer · research systems
Systems for when the demo is over
LLM products, evaluation harnesses and research tooling — built around the data, edge cases and operating constraints that decide whether they hold up
Code, research notes, and the parts that broke
Selected projects
A local AI evaluator and a real-time system — two different ways of making engineering claims inspectable.
All projects01 · Local AI · evaluation
A local AI reviewer for the awkward moment when a polished product listing and the file that arrived may not be telling the same story.
It separates directly measured facts from LLM interpretation — and leaves uncertain claims unverified.
02 · Real-time systems · Web3
Real-time board · settlement ledger · Base Sepolia
A testnet game built as a real-time systems exercise: graphics, state, ledger and contract settlement have to agree.
Live Socket.IO state, a PixiJS board, EIP-712 settlement and a double-entry ledger.
Latest research
A live ML pipeline compared individual trade sizes with a per-second NAV threshold. large_share collapsed toward zero; the audit found three more train/serve feature-contract bugs.
Read the case study →
An HFM exit rule reported +25.98 bps per event. An audit found session concentration, reused validation, and missing joint outputs. A practical research postmortem.
A RAG audit found that source deduplication also changed the evidence budget. A fixed-budget control corrected the claim; a 15-chain pilot validates the pipeline, not effectiveness.
