Back to Projects
Featured
RAG-Based AI Assistant for Domain-Specific Knowledge
Verifiable AI Chatbot for Healthcare, Legal, Education, and Research Domains
An AI-powered chatbot using Retrieval-Augmented Generation (RAG) to deliver accurate, source-grounded answers across specialized domains.
Mar 2024 - May 2024
2.5 months
Solo Project
CompletedOverview
A full-stack AI chatbot system integrating semantic search and generative models to solve the problem of hallucination in general-purpose LLMs. Key capabilities include:
- Retrieval-Augmented Generation with ChromaDB and Gemini API
- Domain-based document parsing (JSON, PDF, XML)
- Relevance scoring via Cohere embeddings
- User credit management using Upstash Redis
- React-based real-time single-page UI
- Multi-domain document handling with per-query document traceability
- Embedded testing utilities and modular backend design
Problem
General-purpose LLMs hallucinate and lack trustworthiness in high-stakes domains. This project ensures factual grounding in verified, domain-specific sources.
What I Did
- •Designed and implemented FastAPI backend with vector search integration
- •Built React frontend with real-time response handling and usage tracking
- •Connected Gemini API with contextual document chunk injection
- •Optimized chunking, embedding, and indexing pipelines
- •Set up environment-based configuration and secure API key handling
Key Features
✓RAG Architecture with Chunk-Level Retrieval
✓Modular Embedding System using Cohere
✓Generative LLM Integration with Gemini
✓Real-Time Query-Response Interface (SPA)
✓Document Source Tracking and Highlighting
✓Usage-Based Credit System
✓Multi-format Data Ingestion (PDF, JSON, XML)
✓Upstash Redis for Rate Limiting & Usage Tracking
✓Fully Configurable via .env Setup
✓Responsive UI with Theming Support
Impact
Demoed to stakeholders in education and healthcare sectors for use in policy compliance Q&A bots.
Technologies
PythonFastAPIReactTailwind CSSCohere APIGemini APIChromaDBUpstash RedisJSONPDFdotenvReact ContextVite