title	Docwise API
emoji	📄
colorFrom	blue
colorTo	green
sdk	docker
app_port	7860
pinned	false

DocWise — Multi-Document RAG SaaS

Upload your documents. Ask anything. Get answers with sources.

---

What it does

DocWise lets you upload PDF documents and have a conversation with them using AI. Ask questions across multiple documents at once, compare them side by side, or search semantically across your entire library. Every answer cites its source document.

---

Features

📄 Multi-document upload

Upload multiple PDFs. Each one is chunked, embedded, and indexed into a per-user vector store. Document records persist across sessions via PostgreSQL.

💬 Conversational RAG with memory

Chat with your documents with full conversation memory. Follow-up questions work correctly — "explain that more" and "what did you mean by X" understand context from previous turns.

🔍 Semantic cross-document search

Search for a concept across your entire document library. Results ranked by similarity score with source attribution.

📊 Document comparison

Ask the same question across two or more documents simultaneously. Useful for comparing contracts, research papers, or policy documents.

📍 Source citations

Every answer includes which document it came from. No silent hallucination.

⚡ Async ingestion pipeline

Upload returns immediately. PDF processing runs as a background task. Poll the status endpoint to check when your document is ready to query.

---

Architecture

┌──────────────────────────────────────────────────────┐

│ Streamlit Frontend │

│ Upload · Chat · Search · Compare │

└───────────────────────┬──────────────────────────────┘

│ HTTP

┌───────────────────────▼──────────────────────────────┐

│ FastAPI Backend — HuggingFace Spaces │

│ Upload · Ingest · Chat · Search · Compare │

└──────────┬────────────────────────────┬──────────────┘

│ │

┌──────────▼──────────┐ ┌───────────---▼---─────────────┐

│ PostgreSQL │ │ RAG Layer │

│ Neon (cloud) │ │ LangChain + ChromaDB │

│ Document records │ │ Google Gemini 3.1 Flash Lite │

│ Upload status │ │ HuggingFace Embeddings │

└─────────────────────┘ └──────────────-----────────────┘

Ingestion workflow:

1. User uploads PDF via Streamlit
2. FastAPI saves file and creates DB record with status processing
3. Background task chunks PDF → generates embeddings → stores in ChromaDB
4. DB record updated to ready
5. User polls status endpoint, then starts chatting

---

Tech stack

Layer	Technology
Frontend	Streamlit
Backend	FastAPI
Database	PostgreSQL (Neon)
ORM	SQLAlchemy
AI model	Google Gemini 3.1 Flash Lite
RAG framework	LangChain
Vector store	ChromaDB
Embeddings	HuggingFace sentence-transformers
Backend deploy	HuggingFace Spaces (Docker)
Frontend deploy	Streamlit Cloud

---

API reference

GET /health

Health check.

{ "app": "DocWise", "status": "running", "version": "1.0.0" }

POST /documents/upload

Upload a PDF document and start asynchronous ingestion.

Query Parameters

Parameter	Type	Required	Description
user_id	string	Yes	Unique identifier for the user uploading the document

Request Body

multipart/form-data

Field	Type	Required	Description
file	PDF File	Yes	Document to upload and index

Response

{
  "doc_id": "a1b2c3d4",
  "status": "processing"
}

DELETE /chat/{session_id}

Clear conversation memory for a session.

---

Getting started locally

Prerequisites: Python 3.11+, Neon account from neon.tech

git clone https://github.com/subata24/Docwise
cd Docwise
python -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create .env in the root:

GOOGLE_API_KEY
GOOGLE_MODEL
DATABASE_URL=postgresql://user:pass@localhost/docwise
CHROMA_PERSIST_DIR=./chroma_db
UPLOAD_DIR=./uploads

Run backend:

uvicorn backend.main:app --reload --port 7860
# API docs at http://localhost:7860/docs

Run frontend:

streamlit run frontend/app.py

---

Project structure

Docwise/

├── backend/

│ ├── main.py # FastAPI routes

│ ├── ingest.py # PDF chunking + ChromaDB indexing

│ ├── rag.py # LangChain RAG + conversation memory

│ ├── models.py # SQLAlchemy database models

│ ├── database.py # DB connection setup

│ └── schemas.py # Pydantic request/response shapes

├── frontend/

│ └── app.py # Streamlit UI

├── Dockerfile

├── docker-compose.yml

├── requirements.txt

---

Engineering highlights

• Per-user ChromaDB collections — users only retrieve from their own documents
• MMR retrieval (Maximal Marginal Relevance) returns diverse chunks, not repetitive ones
• ConversationBufferWindowMemory keeps last 6 turns — follow-up questions work correctly
• Async background ingestion — no request timeouts on large PDFs
• Document status tracking in PostgreSQL — client polls until ready
• Source attribution on every answer
• Dockerized and deployed on HuggingFace Spaces free tier — no credit card required

---

---

Author

Subata Khan — AI Engineer & LLM Systems Developer

---

_{Built to make document intelligence accessible — one PDF at a time.}