OpenClawBrain
Memory and learned routing for OpenClaw. It gives repeated workflows durable local memory, keeps retrieval bounded at runtime, and improves route quality in the background from traces and corrections.
Jonathan Gu
I build AI systems that learn from repeated work.
I design and operate agentic systems for code, research, operations, and markets. My current focus is OpenClawBrain, a memory and learned-routing layer for OpenClaw.
Previously ML Engineer at Instacart. PhD Economics, UCLA. BA Statistics & Economics, UC Berkeley.
Current Focus
Selected Systems
GUCLAW
Always-on AI operator for real workflows. Runs 24/7 on a headless Mac Mini via OpenClaw, managing code deployments, monitoring, triage, and sub-agent coordination.
CormorantAI
Biotech intelligence agent turning signals into briefings. Ingests feeds, triages trial events, tracks FDA catalysts, and writes pre-market briefings for a biotech-focused asset manager.
Project Pelican
ML-driven options trading system. Handles ingestion, feature engineering, training, calibration, and autonomous execution in one pipeline.
Bountiful Garden
Neighborhood produce sharing web app. Built with React 19, Express, Postgres, and Cloudflare, with an AI agent managing development workflows.
Background
Instacart
San FranciscoMachine Learning Engineer
- Built item-availability models serving real-time predictions across millions of SKUs.
- Shipped growth experiments and forecasting systems for platforms moving billions in annual GMV.
- PhD Economics — UCLA, 2014–2020
- BA Statistics & Economics — UC Berkeley, 2007–2011
Writing & Research
- Corrected RL Updates (Gu 2016) — the policy-gradient derivation behind OpenClawBrain's route learning. PDF
Contact