AI-Powered Customer Support Chatbot (FastAPI + OpenNLP + Lucene + Kafka + React)

Final Year Project (Cloud Computing / NLP): a production-style customer support chatbot that combines:

• Natural Language Understanding (intent + entities)
• Knowledge Base Search (Lucene fuzzy retrieval for FAQs)
• Conversation Memory (Redis session state)
• Real-time Logging & Analytics (Kafka → consumer → metrics)
• Modern UI (React + Vite build served via Nginx)

---

Project Overview

This chatbot answers common support questions, supports multi-turn order tracking, and provides basic product recommendations. It uses a hybrid approach:

• Deterministic rules for critical flows (order tracking, escalation safety)
• Model-based intent classification + entity extraction (OpenNLP + local fallback)
• Lucene search when confidence/scores are strong enough (to avoid “wrong answers”)

---

Architecture (Simple Explanation)

FastAPI (Backend / Brain)
Receives messages from the frontend, manages session state in Redis, calls the Java NLP/Lucene service, applies decision logic, and returns a structured response.

Java Microservice (Apache OpenNLP + Apache Lucene)
Exposes a REST API that runs intent classification + entity extraction and performs Lucene fuzzy search over the FAQ knowledge base.

Kafka (Logging Pipeline)
Backend publishes each interaction (user message, response, intent, sentiment, language). A consumer stores logs to disk for analytics.

React (Frontend / UI)
Chat interface that calls the backend API and displays response + metadata.

---

Features

• Intent Recognition with confidence thresholds (avoids irrelevant answers)
• Entity Extraction (order IDs, names, locations/products where available)
• Order Tracking Flow
  • “Where is my order?” → asks for order ID
  • “12345 / ORD-12345 / #12345” → returns tracking status
• Product Recommendations (deterministic logic + backend function)
• Sentiment Handling with safe escalation rule (sentiment < -0.6)
• Multi-language Support (message translation in/out when enabled)
• Analytics
  • /analytics endpoint for summary stats
  • Kafka audit stream (topic: chat_logs)

---

Repository Structure (University Requirement)

cc-project-group10/
├── docker-compose.yml
├── .env.example
├── src/
│   ├── backend/      # FastAPI + Kafka producer/consumer + tests
│   ├── frontend/     # React UI (Vite build served by Nginx)
│   └── ai-chatbot/   # Java microservice (OpenNLP + Lucene)
├── config/
├── docs/
├── scripts/
├── results/
└── .github/workflows/

---

Quick Start (Docker Compose)

Prerequisites

• Docker Desktop (with Compose)

Run

git clone <YOUR_REPO_URL>
cd cc-project-group10
docker compose up --build

URLs

• Frontend: http://localhost:3000
• Backend OpenAPI docs: http://localhost:8000/docs
• Backend health: http://localhost:8000/health
• Java NLP/Lucene health: http://localhost:4567/health

---

Docker Services (Ports)

• frontend → 3000:3000
• backend → 8000:8000
• lucene (Java NLP/Lucene) → 4567:4567
• kafka → 9092:9092
• zookeeper → 2181:2181
• redis → 6379:6379

---

API Endpoints

Backend (FastAPI)

• POST /chat
  Request:

  { "message": "where is my order", "user_id": "u1", "lang": "en" }

  Response includes:

  • response (final message)
  • intent, confidence (0–100), sentiment
  • entities (extracted data)
  • session_context (memory state)

• GET /health
  Shows API status + Redis mode + Java service connectivity.

• GET /analytics
  Returns aggregated analytics (safe to call; should never crash).

• GET /metrics
  Simplified metrics payload for monitoring/report marks.

Java NLP/Lucene Service

• POST /search
  Request:

  { "query": "refund policy", "topK": 3 }

  Response includes intent, confidence, entities, and Lucene hits.

• GET /health

---

Testing

From repo root:

docker compose up -d --build
docker compose exec backend pytest -q

---

Artifacts & Deliverables

• Final Technical Report (PDF): View Report (See also Markdown version)
• Presentation Slides: View Presentation
• Demo Video: YouTube Link
• Project Board: GitHub Issues Board
• Release Version: v1.0

---

---

Contributors

• M wajahat

---

License

See LICENSE.