Skip to content
← All projects

Real-Time ML

Apex

Team of 4

Real-time F1 pit-wall strategy simulator that runs 10,000 race simulations in under 0.2 seconds.

0.996

R² accuracy

01 — Problem

Race strategists need to predict how a driver's lap times will evolve as tires degrade, fuel burns off, and track position changes, then simulate thousands of "what if" scenarios — different pit windows, undercuts, safety cars — fast enough to matter in a live race.

02 — Approach

  1. 01

    Two-stage ML pipeline: Ridge regression with degree-2 polynomial features trained on 10 circuits predicts a baseline lap time per track; XGBoost predicts the residual using per-lap features like tire stint, track position, compound, and fuel load.

  2. 02

    Ridge regression chosen deliberately for Stage 1 — only 10 data points would make a tree model overfit.

  3. 03

    Vectorized Monte Carlo engine in NumPy runs 10,000 simulations injecting model residuals and Gaussian noise.

  4. 04

    WebSocket-driven live race state; chaos events (rain, safety cars) triggered from an iPad "steward" view ripple into new simulations instantly.

  5. 05

    Gemini generates a 3-sentence radio call narrating outcomes — purely for presentation, no computation.

03 — Architecture

04 — Results

0.996

R² (5-fold CV)

0.35s

mean absolute error

<0.2s

for 10K simulations

Stack

Next.jsFastAPIWebSocketsRidge RegressionXGBoostNumPyGemini

Next

Tarmac

iOS app that turns flight delays into spontaneous city itineraries, with an AI safety layer that never lets a bad plan reach the user.