Rahinul User

I've built security tooling for AI agents and LLM applications, including warden, a fail-secure provenance firewall that blocks tool calls driven by untrusted data. It includes provenance-validated GRC workflows, signed hash-chained audit logs, REST/MCP integrations, and a Prometheus-instrumented dashboard.

I also built guard, a zero-config Python CLI scanner for leaked API keys and prompt-injection risks, and measure, an on-device LLM evaluation dashboard for faithfulness, relevancy, contextual precision, and hallucination analysis.

My multi-LLM router uses FastAPI to route requests across GPT-4, Claude, and Gemini, reducing inference costs by 60% through complexity routing, real-time tracking, and fallback logic. For L'Oréal's University Datathon, I led an NLP pipeline over 10,000+ social-media comments, delivering 85% accuracy across sentiment analysis, spam detection, and category tagging.

I'm also the sole author of published research on trustworthy AI security detection, training a transformer on Ethereum contract opcodes and testing reproducibility through per-head ablation, attention attribution, and activation patching.