Nishi Chandra

Built and deployed a production-grade RAG-based AI assistant for Vedic astrology insights, serving 700–1,000 daily users. Designed end-to-end pipeline including data ingestion, embeddings, vector search (Qdrant), and response generation.

Implemented LangGraph-based orchestration for multi-step workflows including retrieval, chart calculations, and contextual response generation. Developed session-based memory for personalized multi-turn conversations.

Engineered backend using Python (FastAPI, async APIs) and integrated with frontend for real-time interactions. Optimized latency, token usage, and response quality through caching, prompt tuning, and selective model routing.

Established evaluation pipelines to improve grounding and reduce hallucinations. Managed deployment using Docker and ensured system reliability, scalability, and performance monitoring.