Dan Austin
Principal Software Engineer - Playing around with AI agents and RL.
This profile was put together from Dan's public GitHub portfolio, and they haven't joined Vibetown yet, so reach them through their portfolio. Is this you? Contact us to update or remove it.
About
Dan Austin works in AI engineering and ML & data science, based in London (Microsoft). Principal Software Engineer - Playing around with AI agents and RL. Public projects on GitHub include Multi Agent Coding System (Reached #13 on Stanford's Terminal Bench leaderboard), Terminal Bench Rl (GRPO training code which scales to 32xH100s for long horizon terminal/coding tasks) and Ai Trains Ai (RL-training an AI agent to RL-train AI agents), mostly written in Python.
Projects 6
Multi Agent Coding System
Reached #13 on Stanford's Terminal Bench leaderboard. Orchestrator, explorer & coder agents working together with intelligent context sharing.
github.com- github.com
Terminal Bench Rl
GRPO training code which scales to 32xH100s for long horizon terminal/coding tasks. Base agent is now the top Qwen3 agent on Stanford's TerminalBench leaderboard.
Ai Trains Ai
RL-training an AI agent to RL-train AI agents.
github.comOrca-Agent-RL
Qwen3-14B Orchestrator Agent Reinforcement Learning. **Achieved 160% improvement** on Stanford's TerminalBench
github.comTbench Agentic Data Pipeline
Multi-agent synthetic data generation pipeline capable of generating and validating long horizon terminal/coding tasks for RL training
github.comCalculator Agent Rl
Training an LLM to use a calculator with multi-turn reinforcement learning, achieving a **62% absolute increase in evaluation accuracy**.
github.com