RazorEdge

Race Strategist AI – Real-Time Pit Decision & Driver Consistency Analyzer

Comment

he NASCAR . My inspiration began the day I watched my first NASCAR race live. I was struck by the relentless focus of drivers pushing through 300 laps — the revs, the sweat, the burnouts, the sound of pure determination. That blend of human endurance and machine precision made me wonder: could AI feel the rhythm of a race — the micro-decisions, the drop in grip, the perfect lap that defines victory? This curiosity drove me to build Race Strategist AI, a system that learns from telemetry and predicts when a driver should pit before performance fades.

What it does

RazorEdge turns raw, noisy GR Cup telemetry into actionable race strategy, live.

  • Predicts next-lap performance degradation
  • Flags optimal pit windows before tire fall-off destroys pace
  • Delivers driver consistency scores, degradation curves, and strategy simulations
  • Runs 100% in-browser at https://razoredge.quantedgeai.io (powered by Streamlit + Cloudflare Tunnel)

How I built it

72 +hour sprint. Zero compromises.

  1. Data & Modeling Layer (Python / Pandas / XGBoost / LightGBM)
    • Ingests high-frequency telemetry (speed, throttle, brake, lateral G, steering, RPM)
    • Engineers lap-wise degradation features + statistical outlier rejection
    • Trains ensemble models to predict next-lap drift (threshold > 1.8 s → pit alert)
  2. Interactive Visualization Layer (Streamlit)
    • Real-time driver consistency index
    • Degradation curves with ML vs heuristic pit-window overlays
    • One-click exportable strategy reports (.csv)
  3. Deployment
    • Dockerized + Cloudflare Tunnel → instantly live at razoredge.quantedgeai.io

Challenges I ran into

  • Driver IDs mismatched between provisional results and telemetry → built _driver_norm_key normalization engine
  • Extreme outliers (lap times < 20s or > 200s from sensor glitches) → fixed with robust statistical clipping
  • Noisy lateral-G and brake traces → multi-stage smoothing + physics-aware filterin

Accomplishments that I'm proud of

  1. Built a complete race-intelligence engine — from raw telemetry to live tactical dashboard
  2. Delivered insights that actually change race outcomes
  3. Created the first real-time pit-window predictor for GR Cup using pure telemetry
  4. Modular, production-grade architecture ready for multi-season, multi-circuit scaling
  5. Quantifiable predictions — lap time loss per stint, exact tire wear rates, optimal pit lap ±1 accuracy
  6. Tactical UI so clean even non-technical crew chiefs understand it instantly
  7. Delivered real-time insights that actually matter to drivers & strategists The next-lap predictor, tire-degradation model, pit-window heuristic, and driver-norm analysis work together to produce actionable insights—not just graphs.
  8. Created a unified ML + Simulation pipeline RazorEdge combines machine learning, domain heuristics, and race simulations into a single console that helps answer “what should I do right now?”
  9. Designed a tactical, modern UI for instant understanding Every insight—driver cluster, tire wear, pit window, predicted lap time—is presented clearly so even a non-technical user can interpret it at a glance.
  10. Modular engineering that scales
  11. The codebase is built on a clean architecture:
  12. ingest/ data pipeline
  13. feature_engineer/ transformations
  14. models/ ML components
  15. realtime/ strategy simulation
  16. dashboard/ for UI
  17. Delivered quantifiable predictions — not hype Models provide measurable impact:
  18. Lap time estimates
  19. Tire wear rates
  20. Degradation curves
  21. Pit strategy outcomes This turns telemetry into decision-grade intelligence.

What I learned

Designing consistent data pipelines for noisy, incomplete GR Cup race telemetry.

  • How to wrangle real-world, incomplete, high-frequency motorsport telemetry
  • The art of turning sensor noise into strategic signal
  • Engineering responsive Streamlit dashboards with Cloudflare live tunnels for real-time insight.
  • How to design real-time analytical pipelines for high-frequency telemetry.
  • Methods to normalize heterogeneous driver data, unify inconsistent lap annotations, and merge telemetry with event marker

What's next for RazorEdge & QuantEdge

  1. Real-time track-side telemetry ingestion (live API / websocket feed) The next step is to connect RazorEdge to live data so pit crews can see predictions update lap-by-lap.
  2. Full pit-strategy optimizer (multi-lap Monte Carlo engine) Build a higher-accuracy simulator using probabilistic models to evaluate thousands of race outcomes.
  3. Driver digital twins Train personalized models per driver to understand their degradation patterns, consistency, aggressiveness, and drop-off predictions.
  4. Cross-race learning Expand beyond Barber to other circuits and conditions so RazorEdge evolves into a multi-track, multi-driver intelligence platform.
  5. AI-generated commentary layer Use LLMs to narrate strategy: “Driver 86 is losing 0.21s/lap due to tire fade—pit window optimal in next 2 laps.”
  6. Integration into QuantEdgeAI Tactical Cloud RazorEdge will become part of the broader QuantEdgeAI ecosystem alongside: CipherEdge Forensics Echo-6 Tactical Agent FlowFusion AI Pipelines
  7. Polished SaaS product (2026) A browser-based platform for race teams, drivers, and simulation engineers.

Built With

Share this project:

Updates