Dan Austin

multi-agent-coding-system. Reached #13 on Stanford's Terminal Bench leaderboard. Orchestrator, explorer & coder agents working together with intelligent context sharing.

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.

Orca-Agent-RL. Qwen3-14B Orchestrator Agent Reinforcement Learning. **Achieved 160% improvement** on Stanford's TerminalBench

calculator_agent_rl. Training an LLM to use a calculator with multi-turn reinforcement learning, achieving a **62% absolute increase in evaluation accuracy**.

tbench-agentic-data-pipeline. Multi-agent synthetic data generation pipeline capable of generating and validating long horizon terminal/coding tasks for RL training

agentic_environments. A mini-framework to build agentic LLM environments.

auto_agent_optimiser. An AI Agent that automatically optimises other AI agents.