Skip to content

DEMO.md

Latest commit

 

History

History
225 lines (169 loc) · 6.53 KB

DEMO.md

File metadata and controls

225 lines (169 loc) · 6.53 KB

Demo Instructions

This document provides detailed instructions for demonstrating the RAG Chatbot application.

Demo Preparation

1. Prepare Sample Documents

For the best demo experience, prepare 2-3 sample documents:

Option A: Use sample documents

  • Download a few PDF articles or research papers
  • Create a TXT file with relevant information
  • Examples: Technical documentation, product guides, research papers

Option B: Create custom documents Create a simple TXT file with content like:

About RAG Systems

Retrieval-Augmented Generation (RAG) is a technique that enhances Large Language Models 
by providing them with relevant context from a knowledge base. This allows the model to 
generate more accurate and contextual responses.

Key benefits:
- Improved accuracy
- Source attribution
- Domain-specific knowledge
- Reduced hallucinations

2. Environment Check

# Check backend
curl http://localhost:8000/api/health

# Should return: {"status":"healthy","service":"RAG Chatbot API"}

Demo Script

Part 1: Introduction (1 minute)

Talking Points:

  • "This is a RAG Chatbot that lets you ask questions about your documents"
  • "It provides answers with full source attribution"
  • "Built with Next.js, FastAPI, and LangChain"

Part 2: Document Upload (2 minutes)

Steps:

  1. Navigate to the Upload tab
  2. Click "Choose a file" and select a sample document
  3. Show the file preview with size information
  4. Click "Upload Document"
  5. Point out: Success message showing number of chunks indexed
  6. Explain: "The document is split into chunks and converted to embeddings"

Part 3: Asking Questions (3 minutes)

Steps:

  1. Navigate to the Chat tab

  2. Ask a simple question: "What is this document about?"

  3. Show:

    • The thinking indicator while processing
    • The answer appears with proper formatting
    • Sources panel updates automatically

  4. Click "View X sources" link

  5. Demonstrate:

    • Source cards with relevance scores
    • Document excerpts
    • Page numbers (for PDFs)

  6. Ask a more specific question related to the content

  7. Show: Different sources may be retrieved for different questions

Part 4: Statistics Dashboard (1 minute)

Steps:

  1. Navigate to the Stats tab
  2. Show:
    • Total documents indexed
    • Collection name
    • Embedding model information
  3. Explain: "This shows how many document chunks are in the system"

Part 5: Technical Deep Dive (2 minutes) - Optional

Talking Points:

  • "Let me show you what's happening behind the scenes"
  • Open browser DevTools Network tab
  • Ask another question
  • Show:
    • API request to /api/query
    • Request payload with question
    • Response with answer and sources

Demo Questions by Document Type

For Technical Documentation:

  • "What are the main features?"
  • "How do I install this?"
  • "What are the system requirements?"

For Research Papers:

  • "What is the main finding?"
  • "What methodology was used?"
  • "What are the conclusions?"

For Product Information:

  • "What problem does this solve?"
  • "Who is the target audience?"
  • "What are the key benefits?"

Advanced Demo Features

1. Multi-Document Queries

  • Upload 2-3 related documents
  • Ask a question that requires information from multiple sources
  • Show how sources from different documents are retrieved

2. Source Verification

  • Ask a question
  • Show the answer
  • Open the sources panel
  • Read the actual text from the source
  • Verify the answer matches the source

3. Handling Unknown Questions

  • Ask a question completely unrelated to the documents
  • Show how the system responds with "I don't know"
  • Explain: "This prevents hallucinations"

Common Demo Scenarios

Scenario 1: Customer Support Use Case

Setup: Upload product documentation Demo:

  • "How can I reset my password?"
  • "What are the warranty terms?"
  • Show how support teams can use this

Scenario 2: Research Assistant

Setup: Upload research papers Demo:

  • "What were the key findings?"
  • "Compare the methodologies"
  • Show how researchers can quickly extract information

Scenario 3: Internal Knowledge Base

Setup: Upload company policies Demo:

  • "What is the vacation policy?"
  • "How do I submit expenses?"
  • Show employee self-service use case

Troubleshooting During Demo

If backend is slow:

  • "The system is processing thousands of vectors"
  • "In production, this can be optimized with caching"

If answer is not perfect:

  • "The quality depends on document quality and chunk size"
  • "This can be tuned for specific use cases"

If connection fails:

  • Keep a backup video/screenshots ready
  • Explain the architecture while resolving

Post-Demo Q&A

Expected Questions:

Q: How accurate is it? A: Depends on document quality and relevance. We use semantic search to find the most relevant chunks, and GPT-3.5 generates answers based only on those sources.

Q: Can it work with other languages? A: Yes! Both the embedding model and GPT support multiple languages. You'd just need to adjust the models.

Q: What about data privacy? A: Documents are stored locally in ChromaDB. For production, you can use self-hosted models or ensure compliance with data policies.

Q: How much does it cost? A: Main costs are Google Gemini API calls (input + output tokens). You can use local models to eliminate API costs.

Q: Can it scale? A: Yes! ChromaDB can be deployed in client-server mode, and the backend can be scaled horizontally.

Q: What file types are supported? A: Currently PDF and TXT. Easy to add more with LangChain's document loaders (DOCX, HTML, CSV, etc.).

Tips for a Great Demo

  1. Practice beforehand: Know your sample documents well
  2. Clear data between demos: Use the "Clear All Documents" feature
  3. Prepare for slow responses: Have talking points ready
  4. Show the code: Briefly show the clean, readable codebase
  5. Emphasize practical use cases: Connect features to real-world problems

Recording the Demo

If creating a video demo:

  1. Set up screen recording at 1080p
  2. Close unnecessary tabs and applications
  3. Zoom in on text when showing code or sources
  4. Use a script but sound natural
  5. Add captions for key features
  6. Keep it under 5 minutes for attention span

Demo Checklist

Before starting:

  • Backend running and healthy
  • Frontend running on localhost:3000
  • Sample documents prepared
  • Browser cache cleared
  • DevTools closed (or ready to open)
  • Tested upload and query flow
  • Notes/script ready

Good luck with your demo! 🎬