engagepro-chatbot

EngagePro Chatbot

An AI-powered customer support chatbot developed as part of the end-of-course project for Large Language Model Applications (LLMA) module of Specialist Diploma in Applied Generative AI at Ngee Ann Polytechnic School of InfoComm Technology.

The chatbot answers EngagePro-specific questions using Retrieval-Augmented Generation (RAG) and responds to general or technical questions using information retrieved from Wikipedia before generating a natural language response.

---

Project Overview

The objective of this project is to develop an interactive chatbot capable of engaging in natural language conversations while providing accurate and grounded responses.

The chatbot combines:

• Retrieval-Augmented Generation (RAG) for company knowledge
• Wikipedia retrieval for general and technical knowledge
• LangGraph for workflow orchestration
• Streamlit for the web interface
• OpenAI GPT models for response generation

---

Features

• Interactive Streamlit chat interface
• EngagePro knowledge base using RAG
• Chroma vector database
• OpenAI embeddings
• LangGraph workflow orchestration
• LLM-based routing agent
• Wikipedia integration for general knowledge
• Conversation history using LangGraph MessagesState
• Prompt engineering to minimise hallucinations
• Responsible AI guardrails implemented using a dedicated LangGraph Safety Node.

---

System Architecture

                         User
                           |
                           v
                       main.py
                           |
                           v
                     Streamlit UI
                           |
                           v
                  LangGraph Workflow
                           |
                           v
                Safety Guardrail
                  /             \
                 v               v
       Blocked Response     Routing Agent
                              /       \
                             v         v
                      EngagePro      General
                          |             |
                          v             v
                       ChromaDB     Wikipedia
                          |             |
                          v             v
                    Prompt Builder Prompt Builder
                          \             /
                           \           /
                            v         v
                              GPT-4.1
                                 |
                                 v
                           Final Response

Project Structure

engagepro_chatbot/
│
├── main.py              # Application entry point
├── app.py               # Streamlit user interface
├── config.py            # Project configuration
├── graph/               # LangGraph workflow, state and nodes
├── guardrails/          # Safety classification
├── llm/                 # LLM factory
├── prompts/             # Prompt templates
├── rag/                 # Retrieval-Augmented Generation
├── routing/             # Question routing
├── wiki/                # Wikipedia retrieval
├── scripts/             # Development and testing scripts
├── tests/               # Unit tests
├── docs/                # Documentation
├── data/                # Brochure and vector database
└── requirements.txt

---

Technology Stack

Component	Technology
Programming Language	Python 3.11
User Interface	Streamlit
LLM	OpenAI GPT-4.1 / GPT-4o-mini
Framework	LangChain
Workflow	LangGraph
Vector Database	ChromaDB
Embeddings	text-embedding-3-small
Document Loader	PyPDFLoader
Knowledge Source	EngagePro Brochure (PDF)
General Knowledge	Wikipedia REST API
Version Control	Git & GitHub

---

Screenshots

Video demo

Example Questions

EngagePro Questions (RAG)

• What is EngagePro’s mission?
• What products does EngagePro provide?

General Questions (Wikipedia)

• What is Artificial Intelligence?
• Explain Retrieval-Augmented Generation.
• What is Prompt Engineering?

---

Development Journey

The project was developed incrementally with ChatGPT acting as a technical mentor and learning companion throughout the software development lifecycle. Each iteration focused on a single component, allowing features to be implemented, tested, and integrated systematically.

1. Assignment Analysis
2. System Architecture
3. Environment Setup
4. Streamlit MVP
5. PDF Loading
6. Document Chunking
7. Knowledge Base Construction
8. Retrieval-Augmented Generation
9. Routing Agent
10. Conditional Workflow Routing
11. Wikipedia Integration

Each iteration was tested independently before integration into the complete system.

---

Future Improvements

Potential enhancements include:

• Multi-document knowledge base
• Conversation summarisation
• Citation of retrieved sources
• Support for additional knowledge sources
• User authentication
• Chat history persistence
• Docker deployment

---

Author

Ng Chee Wee

Large Language Model Applications (LLMA)

Ngee Ann Polytechnic