Football Fan ⚽️
I build things across algorithms, data systems, and machine learning.
- FPL Optimizer: full-stack ML web app predicting player FPL points and the most optimal Fantasy Premier League squads.
- Transformer from Scratch: PyTorch implementation exploring attention and training loops.
- Equity Portfolio Optimizer: Trains signals from market data and 10-K text, then optimizes long/short weights with risk caps, turnover limits, and performance tracking.
- Neuroscience data pipeline: automated extraction and visualization for behavioral + eye-tracking data.
Python, PyTorch, scikit-learn, XGBoost, Flask, React, Tailwind, PostgreSQL, MATLAB, Bash
- Currently building a compiler from scratch.
- Refactoring the FPL model into a clean scikit-learn/XGBoost pipeline.
- Comparing sequence models to boosted trees on time-series player data.
- FPL Optimizer - repo: https://github.com/Legend112358/your-fpl-repo
- Transformer from Scratch - repo: https://github.com/Legend112358/transformer-from-scratch
- Equity Portfolio Optimizer - repo: https://github.com/Legend11235/equity-portfolio-optimizer
- LinkedIn: https://www.linkedin.com/in/sohumguha/
- Email: sohum.guha@mail.mcgill.ca