Projects
These are side projects I’ve built outside work. Both the code and the write-ups are public.
Expected Points for Fantasy Premier League
A machine learning sandbox that forecasts expected points (xP) for every Fantasy Premier League player and optimises squad selection: starting XI, captaincy, transfers and chip timing, with Monte Carlo simulation to show the range of likely outcomes.
Live site →
Source on GitHub
Race Results, Analysis & Forecasts for Formula 1
A Formula 1 strategy and forecasting site that tracks lap-by-lap race pace, pit stops and tyre degradation, and forecasts pole, podium and win chances for the next race, with title odds drawn from 10,000 simulated seasons.
Live site →
Source on GitHub
More about xP-FPL
xP-FPL is my personal FPL “assistant manager”. It predicts how many points each player is likely to score, then uses those forecasts to plan the team week to week.
What it does
- Team selection: picks the formation, starting XI, bench order and captain.
- Transfer advice: checks whether a transfer is worth making, including whether a −4 hit pays off.
- Chip strategy: suggests when to play the Wildcard, Free Hit, Triple Captain and Bench Boost.
- Risk: runs Monte Carlo simulations to give a range of likely scores, not just one number.
How it works
- Collects historical data and live data from the FPL API.
- Builds features such as rolling form, per-90 rates and team strength ratings.
- Predicts expected points with interchangeable models (MLP, GRU, LightGBM and a component-based ensemble).
- Optimises the squad with integer linear programming (PuLP).
- Simulates thousands of gameweeks to measure risk.
Forecasts, results and past performance are published at amparaj.github.io/xpfpl. The code is at github.com/amparaj/xpfpl.
More about F1 Stratbox
F1 Stratbox is my Formula 1 strategy desk. It turns live timing and historical data into race analysis and forecasts, so I can see why a race played out the way it did and what’s likely to happen next.
What it does
- Standings and title odds: tracks Grand Prix and Sprint points separately, and simulates 10,000 seasons to estimate each driver’s title chances.
- Race analysis: breaks down results lap by lap, including pit stop strategies, tyre degradation and where the “tyre cliff” hits.
- Qualifying: compares sessions sector by sector.
- Forecasts: gives pole, podium and win probabilities for the next race.
- History: covers every F1 season since 1950.
How it works
- Collects lap timing, pit stops, race control and weather data from the OpenF1 API, plus calendar and historical data from FastF1 and Jolpica.
- Adds weather forecasts from Open-Meteo.
- Refreshes the data with GitHub Actions every 10 minutes across race weekends.
- Publishes the results on a React + Vite website, with a local Streamlit dashboard for post-race reviews and a strategy sandbox to test pit stop timing and tyre choices.
Race results, analysis and forecasts are published at amparaj.github.io/f1-stratbox. The code is at github.com/amparaj/f1-stratbox.