CAREERS

Founding AI Engineer

Voltaire is building the business decision engine. Know your next move. We are hiring the founding engineer who delivers it.

Strategy consulting is a $300-400B market that still runs on judgment and slide decks. The Voltaire Engine connects to the systems a company already runs, keeps a knowledge graph of the business, and holds a continuously re-scored ranking of the moves that grow ARR, each with the evidence behind it. Clients see the moves in a dashboard and act on them. Every measured outcome writes back into the graph, so the engine gets sharper with every decision it records.

WHAT YOU WOULD OWN

The engine is four layers: Ingest, Graph, Reason, Work, closed by the evidence loop that writes outcomes back. The unit of the product is the decision object: situation, move, confidence, evidence, test, outcome. Language models read and draft; deterministic code decides; a human disposes of every consequential move. All of it is yours, with three jobs to get done.

Make it real.

A simulated console already sells the vision. You replace the simulation with the live system: a skeptical CEO connects their data and watches the engine rank their real moves, with confidence attached, in minutes. The demo is the sales motion and the investor proof at once.

Make it deployable.

This is the forward-deployed job. Pilots install the engine on a client's revenue data with you embedded: governed read-only connectors into their stack, the decision schema at company scale, durable agent runs that survive approval pauses and process death, outcome windows measured in months. One company or a thousand, same architecture.

Make it trusted.

The engine's word is the product. Confidence stated before the work starts and never adjusted after. Golden sets and regression gates that block a degrading change. Attribution that ties each result to the move that caused it. Inference cost per client tracked weekly. The record of predictions against outcomes is what converts a pilot into an annual contract, and it is the moat.

THE SPLIT

Fabrice owns the market: positioning, pilots, investors. HEC Paris, four startups, $60M raised across ventures, one exit, four years at Logitech. Codes every day, so the conversation stays technical. You own the system. Decisions happen in hours, together.

Stack today: TypeScript, Postgres and pgvector (Supabase), Inngest durable functions, Mastra, Claude, Langfuse. Model-agnostic by design. You will have strong opinions about what stays.

THE DEAL

Founding equity and genuine ownership: you define the engine's architecture and hold a real stake in it. Equity-forward. Cash compensation, location, and remote arrangement open to discussion.

THE BAR

  • 01

    You have worked forward-deployed, or close to it: embedded with a customer, wired their messy real stack, and shipped working software against their data on their timeline.

  • 02

    You have shipped agentic systems that ran unattended in production, with real users or real money, and you can describe what broke when many agents shared state.

  • 03

    You draw a deliberate boundary between deterministic code and model judgment, and you can say why a decision belongs in code.

  • 04

    You treat evaluation as engineering: golden sets, CI gates that block a degrading prompt or model change, live outcome metrics.

  • 05

    You treat inference cost as a first-order concern and have a before-and-after to show.

  • 06

    TypeScript and Postgres fluency, and the temperament to own a system end-to-end.

BEFORE YOU APPLY

Answer these in your application.

  1. 01

    Are you allowed to work in the US, or temporarily in the EU?

    Please do not apply if you are not already located in the US or the EU.

  2. 02

    Have you shipped an agentic or LLM system that ran unattended in production, with real users or real money?

  3. 03

    Have you built evaluation infrastructure (golden sets, CI gates) that blocked a degrading model or prompt change from shipping?

  4. 04

    How many years of professional TypeScript experience do you have?

  5. 05

    Are you comfortable with founding-stage compensation that is equity-forward, with below-market cash until funding closes?

  6. 06

    Have you worked forward-deployed or embedded with a customer, shipping against their live stack and data?

  7. 07

    Have you measurably reduced LLM inference cost in a production system?

HOW TO APPLY

Send the system you are proudest of that ran inside someone else's stack, what broke when you scaled it, and whatever shows the real you (GitHub, X, LinkedIn, Substack). Generic applications get ignored.

Apply