Redwan Rahman

An open-source continuous evaluation framework for RAG pipelines, AI agents, and LLM applications (`pip install nirizan`). Features automated judge-drift attribution, fixed anchor sets with repeatable rescoring, statistical regression gating (Mann-Whitney + Holm-Bonferroni), and CI/CD deployment gates - built on an 8-layer unidirectional architecture.

A research framework for procedural content generation that integrates live hardware telemetry into reinforcement learning reward functions, enabling resource-aware training on consumer hardware. The system extends PPO, A2C, MAML, and RLHF with algorithm-agnostic reward shaping, fast task adaptation, and human preference alignment. It includes a containerized local deployment pipeline, Streamlit dashboard, reproducible experimentation tools, and supports Zelda and Sokoban environments for end-to-end PCGRL research.

A modular astronomy research platform that combines AI-assisted scientific analysis with modern astrophysical workflows. Astro-AI integrates LLM-powered question answering, TF-IDF document retrieval, 21cmFAST cosmic evolution simulations, JWST spectral analysis, and Bagpipes SED fitting into a unified Streamlit application, enabling context-grounded exploration and interpretation of galaxy evolution data.

An Outcome-Based Education (OBE)-first AI learning platform that grounds every response in course materials through hybrid retrieval and curriculum-aware generation. Blended AI combines BM25 and semantic search, CO/PO-aware reranking, LLM-powered question answering, persistent student memory, and learning analytics to deliver explainable, citation-backed, outcome-aligned educational assistance for higher education.