Titanic AI Chabot

Built an AI-powered RAG chatbot that combines semantic search and OpenAI GPT to answer questions using retrieved knowledge base content.

Technologies Used: Python, FastAPI, Streamlit, OpenAI GPT-4.1, ChromaDB, Pydantic, Pytest, REST APIs, JSON, Swagger/OpenAPI

The Titanic AI Chatbot is a Retrieval-Augmented Generation (RAG) application that answers questions about the RMS Titanic using a custom knowledge base, semantic vector search, and OpenAI GPT. The application retrieves relevant Titanic facts from a ChromaDB vector database and provides those facts as context to GPT before generating a response.

The project demonstrates modern AI application architecture by separating the user interface, API layer, retrieval engine, and language model into independent components.

Key Features:

  • Conversational chat interface built with Streamlit

  • FastAPI REST API backend

  • ChromaDB vector database for semantic search

  • Retrieval-Augmented Generation (RAG) architecture

  • GPT-powered natural language responses

  • Automatic Swagger/OpenAPI documentation

  • Health monitoring endpoint

  • Automated testing with pytest

  • Context-aware responses based on retrieved Titanic facts

Architecture:

User Question
↓
Streamlit Frontend
↓
FastAPI Backend
↓
ChromaDB Vector Search
↓
Relevant Titanic Facts Retrieved
↓
OpenAI GPT-4.1 Response Generation
↓
Answer Returned to User

Streamlit Chat

Vector Search

Swagger API

Pytest

Challenges & Solutions

  • Traditional keyword search struggled with different user phrasing → Implemented ChromaDB vector similarity search.

  • Frontend and backend logic were tightly coupled → Separated responsibilities using Streamlit and FastAPI.

  • GPT could generate unsupported answers → Implemented Retrieval-Augmented Generation using retrieved context.

  • API reliability needed verification → Added automated tests using pytest and FastAPI TestClient.

What I Learned

  • Building production-style APIs with FastAPI

  • Creating Retrieval-Augmented Generation (RAG) workflows

  • Using vector databases and semantic search

  • Integrating OpenAI GPT into real-world applications

  • Designing frontend/backend architectures

  • Writing automated API tests with pytest

  • Documenting APIs using Swagger/OpenAPI

Future Improvements

  • Add conversation memory

  • Support larger datasets and multiple document sources

  • Add source citations in responses

  • Deploy using Docker

  • Host on AWS

  • Add user authentication

  • Store chat history in a database

  • Implement hybrid search (keyword + vector search)

Github