My first Robotics Experience
Open to read.
Open to read.

Semideus Cognition is a personal project built on one stubborn rule: a claim of knowledge does not count until typed proof attaches, and self-report never moves state. A technical tour of the vision, the full stack, and the agentic systems behind a 'prove it' learning loop.

I am building Semideus to try answer a simple question. "How humans would do, in a world, where most of the tasks would be done by autonomous agents and machines".

How I built production-ready RL training environments using the Prime + Verifiers ecosystem — covering environment base classes, rubric design, evaluation pipelines, and RL training configuration, with a real fitness AI environment as a hands-on example.

How I designed and built a full-stack, multi-agent AI backend with LangGraph, FastAPI, and Supabase — featuring specialized sub-agents, automated nutrition adaptation, real-time streaming, A2A protocol support, and a complete observability pipeline.

Training and optimizing agentic AI workflows (LangGraph) using Reinforcement Learning techniques like RULER and the ART framework.