PitSense

If you are confused about how we come up with specific values/predictions, our ai assistant can answer those questions.
Web scraped data is stored in csv files which is used to perform predictions

What it does

PitSense is an AI-assisted F1 dashboard that turns live or historical raw telemetry into clear, trustworthy strategy calls in real time.

High-Performance Computing Engine

Runs 100,000+ parallel Monte Carlo simulations across multi-core infrastructure
Processes entire race datasets at 5x real-time speed using distributed computing
Leverages CPU parallelization to deliver sub-second predictions during live races

Intelligent Strategy Translation

Transforms complex tire degradation models, fuel calculations, and position forecasts into simple actionable commands
Provides explainable confidence scores for every recommendation (e.g., "PIT NOW - 87% confidence")
Generates win condition probabilities and lap-by-lap position gain/loss scenarios

Real-Time Decision Interface

Intuitive dashboard displays AI-powered pit timing alerts, optimal strategy windows, and track heatmaps
Integrated chatbot answers critical questions instantly (e.g., "Should we pit now?" or "What's our undercut window?")
Designed for race managers coordinating split-second decisions with drivers at the wheel

Historical Analysis & Insights

Archives entire race datasets with telemetry including car coordinates (X, Y, Z), speed profiles, throttle/brake inputs, and pit status
Performs intensive Bayesian learning computations to extract competitor trends and performance patterns
Surfaces optimal pit windows, tire strategy comparisons, and tactical opportunities that would take analysts hours to uncover manually

Race Engineers: Real-time strategy recommendations during live races
Race Managers: Clear, defensible decisions for driver communication
Strategy Analysts: Historical data mining for competitor analysis and future race preparation
F1 Enthusiasts: Deep insights into race dynamics and team strategies

When scaled, PitSense ensures that petabytes of F1 data become clear, defensible decisions when milliseconds matter most.

Latency vs. Accuracy Tradeoff

Running 100,000+ simulations while maintaining sub-second response times required aggressive optimization of our Monte Carlo engine
Solved by implementing multiprocessing pools and strategic caching of tire degradation models

Interpretability of Complex Models

Raw simulation outputs were too complex for split-second decision-making
Built an Explainable AI layer that ranks decision factors by impact score and translates probabilistic outputs into confidence-scored actions

Real-Time Data Integration

OpenF1 API rate limits and data freshness posed challenges for live race scenarios
Implemented intelligent caching with Bayesian updating to refine predictions as new telemetry arrives without re-running full simulations

Scalability Across Hardware

Monte Carlo performance varied drastically between development machines
Designed adaptive parallelization that automatically detects available CPU cores and distributes workload accordingly

HPC Fundamentals in Practice

Gained hands-on experience with parallel computing architectures, discovering that doubling CPU cores doesn't linearly double performance due to communication overhead
Learned to profile bottlenecks using timing metrics and optimize critical paths in simulation loops

Domain-Specific Model Design

Deep-dived into F1 race physics including tire degradation curves, fuel load impacts (0.03s per kg), and pit stop time loss calculations
Understood that accurate modeling requires balancing physical realism with computational efficiency

Human-Centered AI Design

Realized that powerful algorithms mean nothing if engineers can't interpret results under race pressure
Learned to design for explainability first, using ranked factor lists and natural language summaries instead of raw probability distributions

Real-Time System Architecture

Discovered the importance of progressive updates and confidence intervals when working with streaming data
Implemented Bayesian inference to continuously refine predictions without discarding prior simulations, making the system smarter as races progress