What I'm working on

Building

• Blackreach v5, an autonomous browser agent pushing toward stable. Stealth Playwright, DOM walker, full ReAct loop. Project page
• Rigr, an open-source agent evaluation framework. Frozen baselines, regression detection, compliance-ready audit reports. GitHub
• Bifrost, a research project exploring small-model structured prediction against large general models. Working with historical texts. Related tooling
• Library of Alexandria, a 234K+ file cross-cultural research corpus with RAG pipeline. LlamaIndex, ChromaDB, ongoing ingestion. Project page
• Goldmine, an autonomous trading agent. Cycles on schedule, scores market signals, makes decisions. Paper trading for safety.

Learning

• Small model training methodology — how to make purpose-built models compete with general models on structured tasks.
• Agent evaluation infrastructure — what it takes to know if your agent is actually getting better or worse over time.
• MCP server architecture and agent-to-agent coordination patterns.

Reading

• Papers on speculative decoding and inference optimization for local LLMs.
• Research on undeciphered writing systems: Linear A, Proto-Elamite, Indus script.

Life

Based in Toronto. Working on building out a serious local AI setup piece by piece. Day job in the meantime. Dog named Erza keeps me honest about going outside.

This page

Inspired by nownownow.com. The idea is a snapshot of what I'm actually doing right now, not a polished about page.