Niyaz Nabi

The full system architecture for a virtual-human AI math tutor was designed — everything from the FastAPI inference layer and RAG pipeline to the Unreal Engine client data flow. Response relevance climbed from under 60% to 84% after the HuggingFace + Qdrant semantic retrieval pipeline was tuned, while hallucination rates were cut nearly in half.

Over two years, production REST APIs and React/Next.js interfaces were shipped for a SaaS hiring platform serving 100+ enterprise clients — maintaining 99.9% uptime with zero post-release incidents across every deployment. The work most valued was the PostgreSQL query optimization pass that recovered 35% of lost API performance by profiling and re-indexing candidate search queries.