Skip to content

Quick Start Guide

Get up and running with Ollama PDF RAG in 5 minutes.

Prerequisites

  • Completed the installation
  • Ollama running with models pulled
  • Virtual environment activated

The Next.js interface is the modern, feature-rich option.

Start Both Services

# Easy way - uses the start script
./start_all.sh

# Or manually in two terminals:
# Terminal 1 - Backend
python run_api.py

# Terminal 2 - Frontend
cd web-ui && pnpm dev

Open the App

Navigate to http://localhost:3000

Next.js Interface

Workflow

1. Upload PDF     →  Click 📎 button, select PDF
2. Select PDF     →  Check ☑️ the PDF in sidebar
3. Ask Question   →  Type in chat box, press Enter
4. Get Answer     →  Response with source citations

Option 2: Streamlit UI (Classic)

The Streamlit interface is simpler and good for quick testing.

Start the App

python run.py

Open the App

Navigate to http://localhost:8501

Streamlit Interface

Basic Usage

1. Upload a PDF

Next.js: - Click the 📎 (paperclip) button in the chat input - Select your PDF file - Wait for processing (shown in toast notification)

Streamlit: - Use the file uploader in the sidebar - Or click "Load Sample PDF"

2. Select PDFs for Context

Next.js: 

☐ Warranty_Book.pdf (11 chunks • 1 pages)
☑️ Security_Guide.pdf (45 chunks • 12 pages)

  • Check the boxes next to PDFs you want to search
  • Use "All" or "None" buttons for quick selection
  • Selection persists across page refreshes

Important

If you don't select any PDFs and ask a document question, you'll see a warning prompting you to select documents.

3. Choose a Model

Click the model selector in the chat input:

Model Best For
llama3.2 Fast, general purpose
qwen3:8b Detailed reasoning, thinking mode
mistral Balanced performance

4. Ask Questions

Good questions: 

✅ "What are the main security recommendations?"
✅ "Summarize section 3 of the document"
✅ "What does the warranty cover?"
✅ "Compare the approaches mentioned in pages 5-7"

The system handles: 

📄 Document questions → Uses RAG with selected PDFs
💬 General questions → Direct chat (no RAG needed)
⚠️ Doc questions without PDFs → Shows warning

5. Understanding Responses

Responses include:

  • Answer: The main response
  • Sources: Which PDF chunks were used
  • Reasoning (qwen3/deepseek): Chain-of-thought process

Example response: 

The document covers three main security aspects:

1. **Authentication** - Multi-factor authentication is required...
2. **Authorization** - Role-based access control using...
3. **Encryption** - All data at rest uses AES-256...

**Sources:**
- Security_Guide.pdf (chunk 3)
- Security_Guide.pdf (chunk 7)
- Security_Guide.pdf (chunk 12)

Service URLs

Service URL Purpose
Next.js UI http://localhost:3000 Modern chat interface
Streamlit UI http://localhost:8501 Classic interface
FastAPI Backend http://localhost:8001 REST API
API Docs http://localhost:8001/docs Swagger UI
Health Check http://localhost:8001/health Status endpoint

Example Session

Here's a complete example workflow:

# 1. Start services
./start_all.sh

# 2. Open browser to http://localhost:3000

# 3. Upload a PDF (click 📎)
# → Select "company_policy.pdf"
# → Wait for "uploaded successfully!" toast

# 4. Select the PDF (check ☑️ in sidebar)
# → Shows: "Documents (1/1)"

# 5. Ask a question
# → "What is the vacation policy?"

# 6. Get answer with sources
# → "According to the document, employees receive..."
# → Sources: company_policy.pdf (chunk 5, 8)

# 7. Follow up
# → "What about sick leave?"
# → Uses same document context

Tips for Best Results

Document Preparation

  • Use text-based PDFs (not scanned images)
  • Smaller PDFs process faster
  • Split very large PDFs into sections

Asking Questions

  • Be specific: "What is X?" vs "Tell me about X"
  • Reference sections: "In the introduction..."
  • Ask follow-ups: "Can you explain more about...?"

Performance

  • Select only needed PDFs (fewer = faster)
  • Use llama3.2 for speed, qwen3:8b for detail
  • Clear chat history periodically

Troubleshooting

"No PDFs uploaded yet"

→ Click 📎 and upload a PDF first

"No documents selected"

→ Check the ☑️ boxes next to PDFs in sidebar

Response is slow

→ Try a smaller model or select fewer PDFs

"Model not found"

→ Run ollama pull <model-name>

Next Steps