Designed a stateful, multi-turn AI system with long-term memory (PgVector), structured output parsing, and function-calling tools - delivering context-aware career guidance via voice and text at production scale.
Built a real-time LLM evaluation engine that scores project submissions against structured rubrics and generates retrieval-grounded, actionable feedback with consistent output format via JSON schema enforcement.
Implemented automated assessment pipeline generating personalised learning paths with predictive analytics and natural language generation, reducing curriculum design effort by 70%.
Stack: FastAPI � LangChain � React � Voice AI (ElevenLabs) � PgVector � PostgreSQL � AWS
