🚗 AutoCare

AI-Powered Semantic FAQ Chatbot & Knowledge Management System

Enterprise-grade customer support automation with semantic search, real-time FAQ matching, and intelligent department routing — built without heavy ML frameworks.

Documentation · Live Demo · Admin Portal · API Docs

---

📐 System Architecture

graph LR
    A["🧑 Customer"] -->|Query| B["Frontend<br/>index.html"]
    B -->|POST /chat| C["FastAPI<br/>Backend"]
    C -->|Encode| D["FastEmbed<br/>all-MiniLM-L6-v2"]
    D -->|Cosine Similarity| E["FAQ Vector Store<br/>numpy"]
    E -->|Best Match| C
    C -->|Response| B

    F["🔐 Admin"] -->|Manage| G["Admin Portal<br/>admin.html"]
    G -->|CRUD + PDF Upload| C
    C -->|Read/Write| H["faqs.json"]
    C -->|Log| I["escalations.json"]
    C -->|Analytics| J["unanswered_queries.log"]

    style A fill:#4A90D9,stroke:#333,color:#fff
    style F fill:#E74C3C,stroke:#333,color:#fff
    style C fill:#009688,stroke:#333,color:#fff
    style D fill:#FF6F00,stroke:#333,color:#fff
    style E fill:#8E24AA,stroke:#333,color:#fff

Loading

⚡ Key Features

🧠 Semantic Search Vector similarity matching using FastEmbed ONNX (all-MiniLM-L6-v2). Handles paraphrases and natural language queries without keyword dependency.

🏷️ Smart Routing Auto-classifies incoming queries into Service & Maintenance, Insurance, Spare Parts, Roadside Assistance, or General departments.

📄 PDF Ingestion Upload PDF documents to auto-extract Q&A pairs using multi-strategy parsing (tagged, heuristic, fallback). Instantly indexed and searchable.

📊 Gap Analytics Logs unanswered queries and negative feedback to identify knowledge base gaps. Admin dashboard for real-time review and escalation tracking.

🛠️ Tech Stack

Layer	Technology	Purpose
Backend	FastAPI + Uvicorn	Async REST API server
Embeddings	FastEmbed (ONNX Runtime)	Lightweight semantic encoding (~100MB RAM)
Model	all-MiniLM-L6-v2	384-dim sentence embeddings
Vector Ops	NumPy	Cosine similarity computation
PDF Parser	PyPDF	Document text extraction
Frontend	Vanilla HTML/CSS/JS	Zero-dependency UI
Auth	Bearer Token (HTTPBearer)	Admin endpoint protection

🚀 Quick Start

# Clone
git clone https://github.com/shivanshi-git/AutoCare.git
cd AutoCare

Terminal 1 — Backend:

cd backend
pip install -r requirements.txt
python main.py
# → http://127.0.0.1:8000

Terminal 2 — Frontend:

cd frontend
python -m http.server 3000
# → http://127.0.0.1:3000

One-liner (PowerShell):

Start-Process cmd -ArgumentList '/k cd /d backend && pip install -r requirements.txt && python main.py'
Start-Process cmd -ArgumentList '/k cd /d frontend && python -m http.server 3000'

🌐 Live Demo

Interface	URL
Customer Chatbot	autocare-frontend.onrender.com
Admin Dashboard	autocare-frontend.onrender.com/admin.html
API Swagger Docs	http://127.0.0.1:8000/docs (local)

Admin API Key: admin-secret-key

📡 API Reference

Method	Endpoint	Auth	Description
POST	/chat	✗	Semantic FAQ matching — returns best answer + confidence
GET	/api/qa	✗	List all FAQ entries
POST	/api/qa	✓	Add new Q&A pair (auto-embeds & indexes)
DELETE	/api/qa	✓	Remove Q&A pair by question text
POST	/api/upload-pdf	✓	Extract & ingest Q&A pairs from PDF
POST	/api/escalation	✗	Submit customer escalation ticket
GET	/api/escalation	✓	List all escalation tickets
PUT	/api/escalation	✓	Update ticket status
DELETE	/api/escalation/{id}	✓	Delete escalation ticket
GET	/api/unanswered	✓	View unanswered query gaps
POST	/api/feedback	✗	Submit thumbs up/down feedback
GET	/api/verify	✓	Validate admin API key

📂 Project Structure

AutoCare/
├── backend/
│   ├── main.py                 # FastAPI app — routes, embeddings, matching
│   ├── faqs.json               # Knowledge base (55+ Q&A pairs)
│   ├── escalations.json        # Customer escalation tickets
│   ├── unanswered_queries.log  # Gap analytics log
│   └── requirements.txt        # Python dependencies
└── frontend/
    ├── index.html              # Customer chat interface
    ├── admin.html              # Admin management dashboard
    ├── script.js               # API client & UI logic
    └── style.css               # Responsive styling

🔄 How It Works

User Query → FastEmbed Encode → Cosine Similarity vs FAQ Embeddings
                                         ↓
                              Score ≥ 0.45 → Return matched FAQ answer
                              Score < 0.45 → Log to gaps + fallback message

1. User submits a query via the chat interface
2. Backend encodes the query into a 384-dim vector using FastEmbed
3. Computes cosine similarity against pre-computed FAQ embeddings
4. Returns the best match if confidence ≥ 45%, otherwise logs it as a knowledge gap
5. Admin reviews gaps and adds missing Q&A pairs via the dashboard

🤝 Contributing

Pull requests welcome. For major changes, open an issue first.

📄 License

MIT