Sql, Pytorch, Python, self-refine tool, LLaMA-Factory, vLLM, RAG retrieval systems, context engineering, Harness design, reward model design, MCP, regression comparison pipelines, training and inference frameworks, Verification, post-training of LLMs, FSDP, uncertainty estimation, error attribution, evaluation-guided Agent optimization, Agent reliability analysis, Memory, search code interpreters, benchmark construction, data pipelines, verl, human annotation, LLM Agents, sandboxes