AI/MLLLM IntegrationFull-Stack Application
E2E AI Chatbot — Document Q&A with Local LLM
Offline AI chatbot with 0.5s response time using Redis caching
End-to-end AI chatbot for document Q&A (Oct 2023 – Feb 2024). Local LLM (GPT4All), FastAPI, Gradio, MongoDB, Elasticsearch, Redis. RAG over your PDFs - runs completely offline. Production-ready system with full-text search, chat history caching, and document ingestion pipeline. Reduced response time from 97s to 0.5s with Redis caching.
Tech Stack
PythonFastAPIGradioGPT4AllMongoDBElasticsearchRedisDockerKubernetesNginxGit/GitHub
Technical Highlights
- Local LLM deployment with GPT4All for offline document Q&A without external API dependencies
- RAG (Retrieval-Augmented Generation) pipeline with MongoDB for document storage and Elasticsearch for semantic search
- Redis caching layer reducing response time from 97s to 0.5s per question (194x speedup)
- Full-stack architecture: FastAPI backend, Gradio UI, Nginx reverse proxy, Docker containerization, and Kubernetes orchestration
Challenges & Solutions
- Slow inference time → Implemented Redis caching for chat history, achieving 194x speedup (97s → 0.5s)
- Document ingestion at scale → Built ETL pipeline with Logstash for MongoDB to Elasticsearch data migration
Gallery