2025Live
GitaGuru AI
Python, LangChain, LangGraph, CrewAI, Llama 3.1, Claude API, Qdrant, Docker
Personalized Spiritual Learning Platform with production RAG, 6-agent system (CrewAI), and adaptive learning algorithm.
- Built production RAG system with fine-tuned Llama 3.1 8B (QLoRA), hybrid retrieval (BM25 + vector), and cross-encoder reranking on 700+ Bhagavad Gita verses, achieving 92% answer relevance and <2s response latency using vLLM inference.
- Designed 6-agent system (CrewAI) generating personalized spiritual learning curricula: assessment agent profiles users via conversational extraction (89% accuracy), verse selector leverages RAG with metadata filtering, modern application agent (Claude API) contextualizes ancient wisdom for contemporary scenarios.
- Implemented adaptive learning algorithm analyzing user reflections through sentiment analysis (DistilBERT), automatically adjusting curriculum difficulty and verse sequencing, reducing dropout rate by 45% in beta testing.
- Deployed scalable architecture with FastAPI, Gradio UI, SQLAlchemy ORM, Redis caching, and Docker containerization; integrated monitoring via Prometheus and LangSmith tracking 15+ quality/performance KPIs.