Chess Coach Copilot

Inspiration

I love playing online chess and solving chess puzzles. It's a great way to improve logical abilities. I wanted to create a tool that could help other chess players learn and improve their game.

What it does

The Chess Coach Copilot answers chess-related questions about openings, tactics, and other aspects of the game. It provides accurate and precise answers using a combination of vector search and GPT-4. The copilot also features an interactive chess board that displays positions in real-time as the AI discusses them.

How we built it

Embeddings & Storage: Generated embeddings from a collection of great chess books and stored them in MongoDB.

Question Handling: When a user asks a question:

A vector search finds the most relevant answer from the embeddings.
The result is sent to GPT-4 to generate a precise response.

Tech Stack:

Frontend: React with Docker
Backend: Node.js with Express and Docker
Database: MongoDB
Deployment: Azure services for Docker containers and web hosting
Interactive Chess Board: Integrated a chess board to visualize positions.

Challenges we ran into

GPT-4 Response Time: Initially, GPT-4 requests took over a minute to process. Solution:

Optimized vector search to return the single most relevant result.
Combined the query result and GPT-4 response for faster processing.
Implemented a character limit on prompts to avoid sending large requests.

Accomplishments that we're proud of

Interactive Chess Board: Real-time position display is a key feature.
Response Time Optimization: Successfully reduced GPT-4 processing time to under 30 seconds.