Posts by Collection

GitHub Repository

• Designed a multi-stage agentic framework that autonomously completes the ML lifecycle using local LLMs.
• Implemented an execution-feedback decision loop where the agent iteratively drafts code, analyzes traceback signals, and refines strategies through self-correction in a sandboxed environment.

GitHub Repository

• Imitation Learning: Implemented Flow Matching vs. MSE for Action Chunking in Push-T environments.
• Online RL: Implemented major RL algorithms (e.g. GAE, DQN, SAC) on Atari and MuJoCo tasks.
• LLM RL: Fine-tuned LLMs using GRPO and GR-REINFORCE for complex mathematical reasoning tasks.
• Authored a literature review on In-context RL. [PDF]

GitHub Repository

• Built a self-summarizing research agent for BrowseComp-Plus that iteratively executes tool calls (e.g. search, retrieve docs) while automatically compacting interaction history to operate within a fixed context window.
• Developed an end-to-end reinforcement learning pipeline with vLLM-based rollout generation, reward-aligned trajectory extraction, and GRPO training using VERL.
• Enabled Qwen3.5-9B to solve long-horizon research tasks with context compaction, achieving a 2× improvement in accuracy on BrowseComp-Plus.