talkwithdoc

rag application — upload pdfs, pptx, images and chat with them using azure ai search + azure openai.

how it works

1. upload — drop a pdf, pptx, or image
2. parse — extracts text (pdf-parse for pdfs, pptx-parser for slides, azure form recognizer ocr for scanned docs/images)
3. chunk — splits content into overlapping chunks
4. embed — generates embeddings via azure openai
5. index — stores chunks + vectors in azure ai search
6. query — hybrid search (keyword + semantic + vector) retrieves relevant chunks
7. answer — azure openai generates answers grounded in your documents

tech stack

layer	tech
frontend	react 19, vite, mui, react-router
backend	express 5, node.js
search	azure ai search (hybrid: keyword + semantic + vector)
llm	azure openai
storage	azure blob storage
parsing	pdf-parse, pptx-parser, azure form recognizer (ocr)

architecture

frontend (react :5173)
  ↓
backend (express :5000)
  ├── /api/upload     → parse → chunk → embed → index (azure ai search)
  ├── /api/chat       → hybrid search → azure openai → answer
  └── /api/documents  → list uploaded docs

setup

prerequisites

• node.js 18+
• azure subscription with:
  • azure openai (deployment with embeddings + chat model)
  • azure ai search
  • azure blob storage
  • azure form recognizer (for ocr)

environment variables

# backend/.env
AZURE_OPENAI_ENDPOINT=
AZURE_OPENAI_KEY=
AZURE_OPENAI_DEPLOYMENT=
AZURE_OPENAI_EMBEDDING_DEPLOYMENT=
AZURE_SEARCH_ENDPOINT=
AZURE_SEARCH_KEY=
AZURE_SEARCH_INDEX=
AZURE_STORAGE_CONNECTION_STRING=
AZURE_STORAGE_CONTAINER=
AZURE_FORM_RECOGNIZER_ENDPOINT=
AZURE_FORM_RECOGNIZER_KEY=

backend

cd backend
npm install
node scripts/createSearchIndex.js  # create azure ai search index
npm start                          # runs on :5000

frontend

cd frontend
npm install
npm run dev  # runs on :5173

project structure

backend/
  server.js
  routes/
    upload.js              # file upload + processing pipeline
    chat.js                # rag query endpoint
    documents.js           # list documents
  services/
    pdfParser.js           # pdf text extraction
    pptParser.js           # pptx text extraction
    ocrService.js          # azure form recognizer ocr
    ocrLargePDF.js         # ocr fallback for scanned pdfs
    chunkingService.js     # text chunking with overlap
    embeddingService.js    # azure openai embeddings
    searchIndexer.js       # index chunks into azure ai search
    searchQueryService.js  # hybrid search (keyword + semantic + vector)
    chatService.js         # azure openai chat completion
    blobStorage.js         # azure blob storage
  scripts/
    createSearchIndex.js   # index schema setup
frontend/
  src/
    pages/                 # upload + chat views
    theme/                 # mui theme config

supported formats

• pdf (text-based + scanned with ocr fallback)
• pptx (powerpoint slides)
• images (png, jpg — via azure form recognizer ocr)

license

MIT