• 

  Charts Analysis

• 

  Telemetry Health

• 

  The homepage UI

• 

  Race Summary

🏎️ Project Story – RaceSense: COTA Real-Time Strategy Engine

🌟 About the Project

---

🎯 Inspiration

This project was born literally within hours after I found out that the telemetry competition dataset had gone public. At the same time, I’m currently deep in my dissertation, while also being a full-time husband and dad. So the biggest challenge was obvious: limited time, limited energy, but unlimited curiosity.

I’ve always been fascinated by how race engineers make split-second decisions based on messy, imperfect data. When I saw the known issues in this dataset—drifting ECU timestamps, missing lap counts (sometimes even jumping to 32768), inconsistent car IDs—I instantly felt: “Okay, this is the perfect playground for data engineering and real-time reasoning.”

---

🏗️ How I Built the Project

The project was built through a fast but structured approach:

1. Understanding the Telemetry Chaos

COTA telemetry comes with several issues:

• Inaccurate ECU timestamps

• Missing or corrupted lap numbers

• Car numbers assigned inconsistently

• Lap and telemetry files not always aligned

So I started by finding the most stable signals: speed, longitudinal acceleration, GPS traces—and used them as anchors to reconstruct reliable lap boundaries.

1. Constructing a Clean Lap Model

To build the “RaceSense Engine,” I set up a pipeline:

a. Timestamp Normalization Deriving a more trustworthy time axis using meta_time (receiver time) + drift compensation.

b. Lap Boundary Reconstruction

• Detect lap start via GPS loop closure

• Cross-check with speed troughs + location clustering

• Rebuild lap numbers sequentially and consistently

c. Vehicle Identity Resolver

• Prioritizing chassis number as the stable ID

• Automatically detecting updated car numbers across sessions

1. AI-Powered Racing Insights

Once the data was clean, I built an analysis layer that can generate:

• Pace delta analytics

• Tire degradation estimation

• Pit window optimization

• Fuel burn modeling

• Sector-level microstrategy insights

This project focuses on the core engine, not a full race simulator. The priority is simple: 👉 Clean noisy telemetry, reconstruct laps accurately, and extract real-time strategy insights.

---

🧠 What I Learned

• Motorsport telemetry is extremely dirty by nature—and that’s what makes it fun.

• Data engineering matters far more than fancy modeling—garbage in, garbage out.

• Race analytics must be fast and precise, just like the sport itself.

• Even with limited time, focusing on core value produces meaningful results.

---

🧩 Challenges

• Very tight timeline. I worked on this in micro-gaps between dissertation revisions, family duties, and everything in between.

• Dataset inconsistency. Timestamp drift and corrupted lap counts made reconstruction more complex than expected.

• Balancing ideal vs practical. Building a full race model wasn’t realistic in such a short time, so I focused on the most impactful components.

---

🚀 Conclusion

RaceSense was built from a mix of:

• love for data,

• passion for motorsport,

• a bit of acute stress,

• and extremely limited time.

The result is a cleaning + reconstruction + strategy engine that can serve as the foundation for advanced racing analytics.

With further development, RaceSense could evolve into:

• a race engineer assistant,

• pit strategy advisor,

• driver training analytics tool,

• or a multi-track real-time telemetry engine.

Built With

• next
• react
• tailwind
• typescript
• vercel