Writing
Long-form write-ups on AI agent design, retrieval, backtesting and trading-system research. Closer to an engineering notebook than a blog.
Notes from building agents that have to work inside real constraints.
I am building an AI agent that debugs parasitic-extraction runs inside a locked-down VM, with no vector database and no embedding model. It follows a 14-phase roadmap, one gated phase at a time. The newest layer tells it which playbook fits a debugging task and which files to open first.
AI engineering
8 min
Three phases of prediction-market research, from building a dataset to recording feeds for HFT, and the decision to stop.
I set out to spot mispriced long-horizon markets on Polymarket. No public dataset had the history I needed, so I built the collector first, and learned that collection, not modelling, was the bottleneck.
Polymarket journey · 1/3
6 min
I built a vectorized backtester with four entry modes and walk-forward selection, then ran a live bot on Polymarket's order book. Once the simulator charged for latency, fees and slippage, expected value stayed near zero, so I stopped.
Polymarket journey · 2/3
8 min read
Second-scale trading research on Polymarket, built recorder-first: Rust on the hot path, Python for datasets and LightGBM. Good classifier metrics did not turn into profit, and a 98% backtest hit rate fell to about 55% with 300 ms of reaction latency.
Polymarket journey · 3/3
8 min
The reusable rules that kept the research honest: realistic backtests and bots as infrastructure.
A condensed playbook for honest short-horizon backtests: execution order inside one bar, reaction latency, slippage, payout bugs, timestamp alignment and leakage audits. They are the rules that kept my Polymarket research honest.
Execution playbook · 1/2
7 min read
Notes from a live 15-minute bot and an HFT stack on Polymarket. The signal is one component; most of the work is state, fills, reconciliation and keeping blocking calls off the hot loop.
Execution playbook · 2/2
7 min