🏁 Your laps, read like a race engineer would.

Pit Crew Coach is an open-source practice coach for iRacing. It records a stint straight from the sim, or opens a telemetry file, and shows you exactly where the time goes: corner by corner, against your best lap, your typical lap, or the quickest cars in the session. Then a race engineer, powered by a language model running on your own PC, tells you about it over the radio.

Open source Β· Rust Β· runs locally on Windows, macOS, and Linux (live recording on Windows)

Pit Crew Coach run overview for Watkins Glen: best lap, optimal lap, consistency, trend, focus sector, and a pace chart

What it does

Telemetry tools show you graphs. Pit Crew Coach tells you which corner to fix and how.

⏱️

Laps and sectors at a glance

Best lap, optimal lap from your best sectors, consistency, trend across the stint, and the one sector where a typical lap loses the most. Laps that went off track or picked up incidents are flagged automatically.

πŸ—ΊοΈ

A track map that explains itself

The map is colored by time gained and lost, speed, or pedal inputs, with every turn numbered. Hover anywhere to see where the comparison lap was at the same moment. Traces for speed, throttle, brake, gear, and steering stay synced to the cursor.

β†ͺ️

Corner by corner

Click a turn for an entry and exit breakdown: brake point, minimum speed, which gear was quicker, grip used, ABS, and understeer or oversteer. Elevation-aware notes catch downhill braking zones, crests, and compressions.

🎯

Grip circle and shift analysis

A g-g plot shows how much of the tyres' grip each lap and corner used, with a built-in guide to reading it. Upshifts are checked against the car's shift light and redline, including time spent on the limiter.

🏎️

Compare against the field

During a live session it rebuilds the laps of every other car from iRacing's per-car data and compares you against the quickest drivers in your class: where they start braking, carry more apex speed, or get on the power earlier.

πŸ“»

A race engineer on the radio

After every lap, a short call names the corner that cost the most, why, and what to try next time. Pull into the pits and you get a fuller stint debrief. It's spoken in a natural voice with a radio effect, and kept light on numbers so you can take it in at speed.

🧠

Ask the coach

Ask about any corner for written feedback grounded in your actual telemetry, or read the "next time out" list: the handful of changes most likely to find time on your next run. Other drivers' laps are saved per track and car class, so the field library grows every session.

πŸ”’

Fully local

The analysis, the language model, and the voice all run on your own machine. No account, no cloud service, and no telemetry ever leaves your PC.

See it in action

Real screens from the app, using an anonymized test session from the repo. Click any screen for full size.

Track map colored by time gained and lost, next to synced delta, speed, throttle, brake, and gear traces
The track map and traces, linked: hover one and the others follow.
Turn 6 breakdown comparing entry and exit against another lap
Click a turn for an entry and exit breakdown against another lap.
Corner by corner table with time delta, minimum speed, brake point, gear, grip use, and balance for each turn
Every corner at a glance: time, apex speed, brake point, gear, grip, balance.
Gears and shifts analysis showing early upshifts per gear and each upshift on the lap
Shift analysis: which upshifts are early, late, or on time.
Next time out suggestions and written coach feedback from a local Llama model
Suggestions for the next run, plus written feedback from a local model.
Radio transcript with a short engineer call after each lap
The radio log: one short, spoken call per lap.

How it's built

A personal project I use to sharpen systems programming, data analysis, and applied AI. Some of the interesting problems along the way:

~10klines of Rust
~4klines of UI JavaScript
3platforms: Windows, macOS, Linux
0cloud services
πŸ“¦

A binary file reader from scratch

iRacing's .ibt telemetry format is read by a small reader I wrote, which pulls only the channels and session fields the app needs, so unusual metadata layouts don't break an import. On Windows, live data comes straight from the sim's shared memory.

πŸ“

Signal processing on real data

Live telemetry has no GPS, so track maps are rebuilt from heading and lap distance; on real data they agree with GPS to within about 5 to 7 m. Corners are detected from curvature, and other cars' speed is derived from how their lap distance changes over time, smoothed and resampled onto a shared distance grid so the comparison is like for like.

πŸ€–

Grounded, local AI

The coach prompts a local model through Ollama with structured findings (corner metrics, history, conditions) rather than raw data, so feedback stays specific and checkable. Text-to-speech runs through Piper or Kokoro in small Python sidecars, with a radio effect applied in Rust.

πŸ›‘οΈ

Privacy-aware test data

Real telemetry files contain names, IDs, and car setups, so they're never committed. A built-in exporter produces anonymized fixtures that drop identifying fields entirely, and the test suite checks committed fixtures for anything identifying.

🧱

Simple, portable architecture

An async Rust server (axum on tokio) serves a local HTTP API and a React UI with no build step. Other drivers and their laps are stored in SQLite with versioned schema migrations. Everything except live capture builds and runs on macOS and Linux, and a replay mode streams a recorded file as if it were live.

RustTokioAxumSQLiteReactCustom chartsOllamaPiperKokoroPythonuv

Run it yourself

You'll need a Rust toolchain. Ollama and a voice model are optional extras.

git clone https://github.com/islerm/iracing-pitcrew-coach.git
cd iracing-pitcrew-coach
cargo run --release
# then open http://127.0.0.1:8787

# no iRacing? replay a bundled session as if it were live
cargo run --release -- --replay tests/fixtures/roadatlanta-full.ibt --replay-speed 8

Full setup, options, and project layout are in the README. Found a bug or have an idea? Open an issue on the repo.

Projects

What I'm building. More on the way.

🌿

Planthology

An iOS planner that unifies everyday scheduling and trip planning, with drive-time intelligence, encrypted trip sharing, and shareable month recaps.

Learn more β†’

❄️

Snowball Budget

A calm monthly money planner: budget in the style that fits you; zero-based, 50/30/20, kakeibo, or envelopes; log expenses in two taps, grow a savings jar, and snowball your debt to a real debt-free date. No accounts, no servers; everything stays on your device.

Learn more β†’

🌫️

FibroFog

A fibromyalgia tracker built for bad days: one-tap and voice logging, passive weather and health data, flare and pattern detection, and a one-page doctor report. Built in Flutter; Android coming soon.

Learn more β†’

🏁

Pit Crew Coach This project

An open-source iRacing coach in Rust: live telemetry, corner-by-corner analysis, field comparisons, and a local AI race engineer on the radio.

Learn more ↑