vdoc_rag / TESTING_GUIDE.md
aditya9128
Remove redundant README_HF.md
f6fc558

πŸ§ͺ VDoc-RAG Testing Guide

A step-by-step guide to test the VDoc-RAG document Q&A application.


πŸ“‹ Table of Contents

  1. Access the Application
  2. Upload a Document
  3. Ask Questions
  4. Review the Response
  5. Provide Feedback
  6. Additional Features
  7. Troubleshooting

1. Access the Application

Online (HuggingFace Spaces)

Open the live demo: https://huggingface.co/spaces/adi9128/vdoc_rag

Local Development

# Start the server
uvicorn app.main:app --host 127.0.0.1 --port 8000

# Open in browser
http://127.0.0.1:8000

2. Upload a Document

Step 2.1: Select a File

  • Click the "Choose File" button
  • Select a PDF file from your computer
  • Supported formats: .pdf, .png, .jpg, .jpeg

Step 2.2: Upload & Index

  • Click "Upload & Index" button
  • Wait for processing (usually 5-15 seconds)
  • You'll see: βœ… Uploaded and indexed [filename] (X chunks)

Step 2.3: Confirmation

  • A green box will appear: "πŸ“Ž Ready to query: [filename]"
  • The document is now indexed and ready for questions

3. Ask Questions

Step 3.1: Enter Your Question

  • Type your question in the text box
  • Example questions:
    • "What is the user's name?"
    • "What skills are mentioned?"
    • "Summarize this document"
    • "What is the contact email?"
    • "List the work experience"

Step 3.2: Submit

  • Click "Ask Question" button
  • Wait for the AI to process (2-5 seconds)

4. Review the Response

The Answer Section Shows:

Component Description
πŸ’‘ Answer AI-generated response based on your document
πŸ“š Sources Which pages/sections were used
πŸ” Retrieved Chunks Raw text segments found in the document
Confidence % How relevant each chunk is to your question

Example Response:

πŸ’‘ Answer:
The user's name is Aditya, as mentioned in the resume header.

πŸ“š Sources:
- Page 1 β†’ Name: Aditya, Email: aditya@example.com...

πŸ” Retrieved Chunks:
[1] Name: Aditya | Confidence: 95.2%
[2] Contact: aditya@example.com | Confidence: 87.1%

5. Provide Feedback

After receiving an answer, you can help improve the system:

Button When to Use
βœ… Correct The answer is accurate and helpful
❌ Incorrect The answer is wrong or irrelevant

Your feedback helps improve future responses!


6. Additional Features

Clear Cache

  • Click "🧹 Clear Cache" to remove cached responses
  • Use when you want fresh answers

Clear Persistent Index

  • Click "πŸ—‘οΈ Clear Persistent Index" to remove all indexed documents
  • Use to start completely fresh

Dashboards

Dashboard URL Purpose
Feedback /feedback_dashboard View feedback statistics
Benchmark /benchmark_dashboard Evaluate embedding models

7. Troubleshooting

"No document found. Please upload a document first."

  • Cause: No document is currently indexed
  • Solution: Upload a PDF file first

"Internal Server Error" after uploading

  • Cause: Server state issue
  • Solution: Refresh the page and try uploading again

Document shows from another user

  • Note: This is a shared demo instance
  • Solution: Upload your own document - it will replace the previous one

Slow response times

  • Cause: First load downloads ML models (~500MB)
  • Solution: Wait for initial setup, subsequent queries are faster

OCR not extracting text properly

  • Cause: Low quality PDF or scanned image
  • Solution: Use higher resolution PDFs or clear text documents

πŸ“ Test Scenarios

Scenario 1: Resume Testing

  1. Upload a resume PDF
  2. Ask: "What is the candidate's name?"
  3. Ask: "What are their skills?"
  4. Ask: "List their work experience"

Scenario 2: Report Testing

  1. Upload a business report PDF
  2. Ask: "What is the main topic?"
  3. Ask: "Summarize the key findings"
  4. Ask: "What recommendations are made?"

Scenario 3: Invoice Testing

  1. Upload an invoice PDF
  2. Ask: "What is the total amount?"
  3. Ask: "What items are listed?"
  4. Ask: "What is the due date?"

βœ… Success Criteria

The application is working correctly if:

  • Document uploads successfully with chunk count
  • Questions return relevant answers
  • Confidence scores are displayed
  • Source attribution shows page numbers
  • Feedback buttons work
  • New uploads replace old documents

πŸ†˜ Getting Help

If you encounter issues:

  1. Check the browser console for errors (F12)
  2. Try refreshing the page
  3. Upload a different PDF to test
  4. Clear the index and try again

Made with ❀️ by Aditya