1. Doc Chatbot

Overview

A conversational assistant that lets users upload any document (PDF, Word, Markdown) and then ask free-form questions. Behind the scenes it uses retrieval-augmented generation (RAG) to find relevant passages and a generative model to craft answers.

Primary Use Cases

• Knowledge workers: Rapidly find clauses in long policies, specs, proposals
• Students: Query lecture notes, research papers
• Customer support reps: Lookup SOPs or troubleshooting guides

Key Features

• File upload + OCR (for scanned PDFs)
• Semantic search over document chunks (vector store)
• Chat UI with follow-up question support
• On-demand summarization of entire sections

Tech Stack

• Frontend: React + TypeScript + Tailwind (chat interface, file picker)
• Backend: FastAPI (Python) for AI endpoints; optional Go microservice for authentication or rate-limiting
• Database: PostgreSQL (user/docs metadata) + FAISS (vector store)
• AI Models:
  • Embeddings: sentence-transformers/all-MiniLM-L6-v2
  • Generator: google/flan-t5-large or similar from Hugging Face

Architecture

1. Upload service ingests document → PyMuPDF/pdfplumber extracts text → chunk + embed → store in FAISS.
2. Query service receives user prompt → embed → FAISS k-NN → retrieve top passages → concat with prompt → generate answer via T5.