self-correcting-agent

A Personal OS Agent that autonomously researches any topic overnight — inspired by Karpathy's AutoResearch. Set a goal, let it loop, wake up to structured markdown reports. No babysitting required.

Agentic Azure OpenAI test harness that drives GPT-6 Astra through a closed-loop build-and-visual-QA cycle — it builds a WebGL kart racing game, screenshots it in headless Chromium, and self-corrects from the diagnostics until it actually renders and drives.

Autonomous LangGraph workspace agent for iterative coding, document engineering, and sandbox tasks via a secure Docker gateway with Telegram ingress and Langfuse telemetry.

4 hand-built AI agents (Research, Coding, Sales, Recruiting) with real plan→act→observe→correct loops — no LangChain/CrewAI/AutoGen, runs free on Gemini + Colab.

This project takes a folder of PDF research papers and transforms them into a studio-quality, multi-speaker podcast. It uses a LangGraph Agentic Swarm to draft and self-correct the script, LlamaIndex Hierarchical Retrieval for deep contextual accuracy, and Kokoro TTS + Pydub for dynamic voice assignment and cinematic audio post-production.