MindEase AI Assistant

MindEase is a domain-focused chatbot for social anxiety support. It uses retrieval-augmented generation (RAG) over your local document knowledge base, supports text/voice/document inputs, and logs real runtime analytics for evaluation.

Features

• Social-anxiety-focused responses with scope filtering
• RAG pipeline using FAISS vector search
• Multimodal input support:
  • Text
  • Voice
  • Document upload (.pdf, .txt, .md)
• Language response policy:
  • English
  • Urdu script
  • Roman Urdu
• Real-run analytics logging and report chart generation

Tech Stack

• Backend: FastAPI
• AI/RAG: LangChain, FAISS, NVIDIA Embeddings, NVIDIA Chat model
• Frontend: HTML, CSS, JavaScript
• Reporting: Python + Matplotlib

Project Structure

.
├── web_app.py                      # Main FastAPI app (active backend)
├── chatbot_ai.py                   # Core AI logic (retrieval, generation, scope/language control)
├── ingest_database.py              # Build FAISS index from PDFs in data/
├── calculate_accuracy_report.py    # Accuracy summary from real analytics logs
├── generate_report_graphs.py       # Real-data graph generation for reports
├── requirements.txt
├── templates/
│   └── index.html                  # Main UI template
├── static/
│   ├── css/style.css               # Active stylesheet used by template
│   ├── js/script.js                # Active frontend logic used by template
│   ├── style.css                   # Legacy stylesheet copy
│   └── script.js                   # Legacy script copy
├── data/
│   └── *.pdf                       # Source knowledge documents
├── faiss_index/
│   ├── index.faiss                 # Vector index file
│   └── index.pkl                   # Index metadata/docstore
├── analytics/
│   └── chat_metrics.csv            # Runtime metrics log (auto-generated)
└── report_graphs/
    ├── grouped_bar_response_time.png
    ├── histogram_response_time.png
    └── evaluation_dataset.csv

Prerequisites

• Python 3.10+ (Windows commands below assume py launcher)
• NVIDIA API key for model and embeddings

Setup

1. Create and activate virtual environment

py -m venv venv
.\venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

3. Create .env file in project root

NVIDIA_API_KEY=your_api_key_here
CHAT_MODEL=meta/llama-3.1-8b-instruct

4. Add your social-anxiety-related PDFs into data/

5. Build vector index

py .\ingest_database.py

Run the App

py .\web_app.py

Open:

• http://127.0.0.1:7860/

How Runtime Flow Works

1. User sends text/voice/document input from web UI.
2. Backend validates input and session context.
3. For document uploads:
   • Text extraction
   • Social-anxiety scope check
   • Out-of-scope docs are rejected with an outside-knowledge message
4. Query is processed with retrieval from FAISS.
5. LLM generates response with language policy.
6. Metrics are logged to analytics/chat_metrics.csv.

Analytics and Reporting

Accuracy summary

py .\calculate_accuracy_report.py

Output includes:

• Total logged chats
• Overall mean accuracy
• Median accuracy
• Accuracy by input type

Report graphs from real run data

py .\generate_report_graphs.py

Generated files:

• report_graphs/grouped_bar_response_time.png
• report_graphs/histogram_response_time.png
• report_graphs/evaluation_dataset.csv

Notes

• Active frontend files are static/js/script.js and static/css/style.css.
• chatbot.py exists as an alternate/legacy path; primary runtime is web_app.py + chatbot_ai.py.

Troubleshooting

• "FAISS index not found": run py .\ingest_database.py first.
• Missing API key error: verify .env contains NVIDIA_API_KEY.
• Empty analytics/graph errors: run some chats first, then run reporting scripts.