Portfolio Assistant - RAG Chatbot
A RAG-powered chatbot backend that answers questions about my professional portfolio using semantic retrieval, vector search, and streaming responses.
Key Highlights
- ▹ Implemented complete RAG pipeline with semantic chunking, vector indexing, and relevance-threshold retrieval
- ▹ Built REST API with JSON and streaming SSE endpoints, rate limiting, and CORS middleware
- ▹ Deployed on Cloud Run with Docker containerization and automated deployment scripts
- ▹ Created knowledge indexing tool with configurable chunking and MD5-based deduplication
- ▹ Achieved 88% fact accuracy through RAG optimization (up from 27%)
Tech Stack
Overview
The Portfolio Assistant is the AI chatbot powering this very website. It uses RAG (Retrieval-Augmented Generation) to answer questions about my professional experience, projects, and skills with high accuracy.
Technical Highlights
RAG Pipeline
Implemented complete retrieval pipeline with semantic chunking, Pinecone vector indexing, FlashRank reranking, and relevance-threshold filtering.
API Design
REST API with both standard JSON and streaming SSE endpoints for real-time responses, plus rate limiting and CORS middleware.
Production Deployment
Docker containerization with Cloud Run deployment, including health checks and automated deployment scripts.
Knowledge Indexing
Custom indexing tool with configurable chunking, dry-run mode, and MD5-based deduplication for efficient updates.
Accuracy Improvements
Through iterative RAG optimization, improved fact accuracy from 27% to 88%, ensuring reliable responses about my professional background.