Product Engineer — Build the AI Pit Wall

We're building PitGPT — an AI race engineer that reads 120 Hz telemetry from racing simulators, understands what happens at 280 km/h into Eau Rouge, and tells you why you're losing three tenths there.

7,000+ racers in Early Access. Backed by Google for Startups and NVIDIA Inception. Pre-seed, small team, AI-first — we build with AI and ship at a speed that reflects it.

We need the person who takes a feature from idea to production before anyone has time to schedule a meeting about it.

What You'll Actually Do

Our small team designed the core architecture — the data pipeline, the native agent, the telemetry stack. You take that architecture and build on it. Fast.

Your week looks like this:

• Redesigning the lap comparison view based on tester feedback — React, Tailwind, Three.js data viz
• Adding a telemetry field to the provider API so the AI can coach on tyre degradation
• Tuning system prompts so PitGPT gives better corner-by-corner advice for LMU GT3 cars
• A sim update changes data formats overnight — you trace the issue through game files, build the mapping, and ship the fix before anyone notices.
• Something breaks — you check logs, find the cause, fix it. Not a thread about it.
• Optimizing data pipeline performance — query tuning, storage efficiency, shaving milliseconds off API responses
• Setting up a preview deployment so testers can try your feature branch

This is not a frontend role. You work across the full stack — React/Next.js, Bun/Hono API, ClickHouse, prompt engineering, and occasionally extending Rust code in the desktop agent with AI assistance.

Tech Stack

• Web app: Next.js 13, React 18, TypeScript, Tailwind, Zustand, Three.js
• Provider API: Bun, Hono, TypeScript, ClickHouse, Protobuf (ts-proto), Zod
• Desktop agent: Rust + Tauri v2 (React/Tailwind UI) — you extend, not rewrite
• AI layer: OpenRouter, system prompt engineering, tool-use patterns
• Data: ClickHouse (columnar, telemetry-optimized), Protobuf + zstd streaming
• Infra: Docker, Kubernetes (GKE), GitLab CI/CD, GCP
• Sims: Shared memory APIs (ACC, LMU, iRacing), game file parsing

Who This Is For

Someone who fixes things before asking permission.

When something breaks, you check logs, trace the request, test hypotheses, and fix it. You don't escalate what you can solve. You don't over-research when the answer is in the error message. We're a 3-person team — everyone debugs, everyone ships, everyone owns their output.

TypeScript is your weapon.

React, Node, Bun — you build both UI and API without switching gears. You're fast. Not "I'll have a PR by Friday" fast. "It's deployed on preview, go test it" fast.

AI is how you work, not what you talk about.

We're an AI-first company — in the product and in how we build it. That means we ship at a pace that would be impossible without AI tooling. Cursor, Claude, Copilot — whatever your stack. You use it to write code, debug, explore unfamiliar codebases, and learn new tech in hours instead of weeks.

But here's the thing: AI is a multiplier, not a replacement for understanding. When the AI hallucinates a wrong API call, you catch it because you understand HTTP. When the generated code has a subtle race condition, you spot it because you know async fundamentals. We don't want someone who can't function without AI — we want someone who's dangerous with it because they understand what's happening underneath.

You dig into things.

You've modded a game. Written a telemetry overlay. Built a Discord bot that scrapes race results. Reverse-engineered a file format for fun. The instinct to open the hood and understand how something works is what matters.

You ship.

We're pre-seed. You'd rather deploy something testable today than architect something beautiful next month.

Strong signals (not required, but you'll ramp faster):

• Sim racing (ACC, LMU, iRacing, or any sim with telemetry)
• Shared memory / memory-mapped file experience
• Rust — even basic. Being able to read our CTO's code and make guided changes with AI is a real plus.
• ClickHouse, Protobuf, or columnar/binary data formats
• DevOps basics — Docker, k8s, CI/CD

This Is Not For You If:

• Your first instinct when something breaks is to ask someone else instead of checking the logs yourself
• You make decisions based on gut feeling when there's data available — we expect you to look at the numbers first
• You wait for a sync call before starting work on something obvious
• You spend more time discussing architecture than shipping features
• You need detailed specs to start working — we give you a problem, not a blueprint
• You optimize for process over output

How We Work

• Self-sufficiency is the baseline. Our CTO is available for architecture decisions and code review — not for routine debugging. When you're stuck on something genuinely hard, you come with "I tried X and Y, here's what I found." That's how a small team moves fast.
• Small team, high ownership. No product managers, no Jira ceremonies. You own features end to end.
• Async-first. Slack + GitLab. Occasional syncs for alignment. EU timezone overlap for daily responsiveness.
• AI-first means we ship fast. We use AI to build the product AND in the product. A feature that used to take a week takes a day. But we also expect you to understand your tools deeply enough to know when the AI is wrong.

The Deal

• Start: Part-time contract (15–25h/week), €30–50/hr depending on experience
• Path to full-time as we close seed round
• Equity: Available for top candidates — in the range of tenths of a percent, milestone-based, conditional. Earned through delivered impact, not promised upfront.
• Tools: Whatever you need. Cursor, AI subscriptions, hardware — we invest in tools, not bureaucracy.
• Remote-first. Prague office available, but results > chair time.

How to Apply

No CVs. No cover letters. Show us you can solve problems.

Please send a challenge in your reply; otherwise, we will not consider your response valid.

Subject: RACEMAKE – Engineer – [Your Name]

1. One thing you built. Link + one sentence on the hardest technical problem you solved there.

2. The Engineering Challenge.

We prepared a short task. It's not a trivia quiz — it tests whether you can read telemetry data and figure out what's happening on track. When completing it, set the coachVoice config to 'pitgpt' — this activates our production coaching tone.

→ https://gist.github.com/743milan/e36760c15ff8b534a9c915b601b6a5b7

Complete it. Send us the result with your code + JSON output. If we see the signal, we'll be in touch within 48 hours.

We're not looking for someone who needs a four-part onboarding before they can push code. We're looking for the engineer who clones the repo, reads the README, and has a question about the architecture — not about how to run bun install.

Build the pit wall. 🏁