Christopher Flynn

Turning ideas into products, and code into impact—building scalable systems that power real-world solutions.

Project Type: AI Application | NLP | Q&A System | Kaggle Project
Tools Used: Python, Google Gemini API (Flash & Embedding models), google-genai SDK, Pandas, NumPy, Scikit-learn

Link to Kaggle Notebook

Objective

Create a functional and reliable question-answering system capable of understanding natural language questions about specific board game rules and providing accurate answers sourced only from the provided rulebook text, utilizing the power of Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG).

Key Features

  • Retrieval-Augmented Generation (RAG) Core:

    • Ingests rulebook text and splits it into manageable semantic chunks.
    • Generates vector embeddings for each chunk using Google’s text-embedding-004 model to capture meaning.
    • Embeds user questions and performs semantic search (via cosine similarity) to find the most relevant rule snippets.
    • Feeds the question and retrieved context to gemini-2.0-flash to generate a final answer grounded in the source text, minimizing hallucinations.

  • Python & GenAI Stack:

    • Built entirely in Python within a Kaggle Notebook environment.
    • Leverages the google-genai SDK (v1.7.0 pattern) for seamless interaction with the Gemini API.
    • Utilizes Pandas and NumPy for data manipulation and numerical operations.

  • Natural Language Q&A:

    • Allows users to ask questions about rules in plain English (e.g., “How do I win?”, “What does the ‘Explore’ action cost?”).
    • Provides answers synthesized by the AI but constrained by the retrieved, relevant rule sections.

  • Focused Notebook Implementation:

    • Designed as a clear demonstration of the RAG pipeline within a self-contained Kaggle Notebook.
    • Focuses on the core AI capabilities rather than complex UI or deployment infrastructure.
    • Code includes error handling and API retry logic for robustness.

Environment & Implementation Highlights

  • Developed and runs entirely within a Kaggle Notebook.
  • Utilizes Kaggle Secrets for secure management of the Google API Key.
  • Demonstrates simulation of Vector Search using Scikit-learn’s cosine_similarity on NumPy arrays.
  • Code is documented with Markdown cells explaining each step of the RAG process.

🔗 View Code on GitHub: GitHub Link
🔗 View Blog Post: Link

Updated: