reasoning-traces

Self-healing code reasoning engine. Detective → QA → Patcher closed loop on SWE-bench, with an RL layer that turns every reasoning trace into DPO training data.

[CVPR 2026] StaR-KVQA: Structured Reasoning Traces for Implicit-Knowledge Visual Question Answering

Research codebase for an IEEE FLLM 2025 paper: retrieval-augmented LLM agent for CTF challenge solving with RAG, graph memory, tool routing, and reasoning traces.

Creating datasets and training code and dependancies.

Failure-aware audit protocol for reasoning-trajectory analysis in HumanEval code generation.