Tech Stack
Python FastAPI MongoDB Milvus Voyage AI Embeddings Gemini / Vertex AI OpenAI LangChain Playwright BeautifulSoup AWS S3 Google Cloud Storage Uvicorn
Project Highlights
- Built an AI customer support backend that answers questions from documents, websites, FAQs, and images.
- Developed a RAG pipeline using Milvus vector search, Voyage embeddings, reranking, and Gemini/OpenAI models.
- Added website FAQ caching and change detection to reuse embeddings and avoid unnecessary re-scraping.
- Built paraphrase clustering to group similar customer questions and generate improved reusable FAQs.
- Implemented intent-based routing for Q&A, workflows, form inputs, tone/sentiment analysis, and support tasks.
