SCRYM
SCRYM is an AI-powered decision support system that transforms how coaches make critical in-game decisions.
Core trick
Trains a multi-output neural network on 4th-down situations to output conversion probability and a go, punt, or field goal recommendation.
Description
SCRYM is an AI-powered decision support system that transforms how coaches make critical
in-game decisions. It combines PyTorch neural networks with Vision Agents to deliver instant, data-driven recommendations.
README excerpt
# NFL AI Coach: The Playbook (4th Down Bot Demo) 🏈 **NFL AI Coach** is a data-driven decision support system designed for the **Gemini 3 SuperHack 2026**. This demo version focuses specifically on the **4th Down Bot**, a critical component that builds the foundation for the full AI coaching suite. > **Note:** An alternate, unmerged version of this project is available in the [`alternate-work-snapshot`](https://github.com/jorgeajimenez/scrym/tree/alternate-work-snapshot) branch. ## 🚀 Demo Focus: The 4th Down Bot The platform uses a multi-output neural network to analyze 4th down situations and recommend the optimal decision: **Go for it**, **Punt**, or **Field Goal**. ### Key Capabilities 1. **Conversion Probability**: Calculates the likelihood of converting on 4th down. 2. **Field Goal Probability**: Estimates the success rate of a field goal attempt from the current distance. 3. **Expected Points (EPA)**: Compares the expected value of each decision to recommend the mathematically optimal choice. 4. **Win Probability Integration**: Contextualizes decisions based on the current score and time remaining. ## 🛠 Tech Stack - **Frontend**: Next.js 15 (App Router, TypeScript, Tailwind CSS) - **Backend**: Python 3.11 (FastAPI, PyTorch, nfl_data_py) - **Data**: NFL Play-by-Play (PBP) via nflverse (2018-2024). - **Models**: PyTorch Feedforward Networks. ## 📂 Project Structure - `backend/`: FastAPI server and ML training logic (4th Down focused). - `frontend/`: Ne