Forward-Deployed Engineer

You take the hardest, least-proven part of an AI bet inside the R&D team of a large technology company and own it end to end. Vague question to production system, nobody between you and the decision.

Apply How we work

Matouš Havlena on stage at a conference in Prague

The role

Forward-deployed means the problem is yours. You sit inside the client’s R&D team, with their engineers and their constraints, from the first unclear brief to a system that runs every day. The same applies to our own product, where you are the client.

The work has a shape. State the real question in one sentence. Design the shortest experiment that settles it. Write the no when the answer is no, with the evidence, so the budget goes to what is real. Build what survives into production, and stay with it.

The problems sit at the frontier. Agent platforms, protocols and inference infrastructure, and the robotics, sensing and real-time systems around them.

What you will do

Take an unclear question from a research or engineering leader and restate it in one sentence that everyone can test.
Design the shortest experiment that settles it. A two-day benchmark, not a two-month feature.
Write the no when the answer is no, with the evidence, and move the budget to what survived.
Build what survived into a production system, end to end and without a spec, with several threads open at once and none dropped.
Run last month’s model or agent release yourself and know exactly where it broke.
Lead technical deep dives with the client’s engineers, and turn “we want a copilot” into “which decisions can we automate and still pass the audit”.

You may be a good fit if

You have shipped production systems that real users depend on, and you have run them afterwards. Prototypes alone do not count.
You have built LLM or agent systems that work outside a demo, with evaluations that measure whether they do, and you know where they break.
You are strong in at least one production language, Python or TypeScript here, and your code is tested, observable and readable by the next person.
You can frame an ambiguous problem for an enterprise R&D team and hold the technical discussion with their engineers and their decision makers.
You own a problem end to end and move between backend, infrastructure and interface when the problem demands it.
You have killed your own work on evidence at least once, and you can tell that story.
You are driven from the inside, with something to prove and in love with the craft. You rebuilt the pipeline nobody asked about because it offended you, and demoed it on Friday.
You solve the problem, not the ticket, and you say “we” about wins. Former colleagues would join you again.

A degree, a particular number of years or a specific list of tools are not what we screen for. The systems you have shipped and the dead ends you have closed are.

A bonus, not a filter

Work inside regulated environments such as healthcare, finance or the public sector, with security review and audit.
Open-source contributions, published writing or conference talks.
Robotics, sensing or real-time systems.
Consulting or client-facing work.

What you get

Ownership of real systems, from an unclear question to production, with nobody between you and the decision.
The frontier as daily work. Agent platforms, protocols and inference infrastructure inside the R&D teams of large technology companies.
Senior peers and a fast learning pace. No layers of management, no meaningless meetings.
Your name on the work, in Production Notes, open-source code and conference talks.
Prague or San Francisco, hybrid or remote, competitive compensation, and a long-term collaboration in mind.
The honest part. Half the job is killing your own work in writing, and if you need shipped features to feel progress, this role will hurt.

How we hire

Two to three rounds. An intro call, a code challenge you do on your own time, and a technical interview with the team leads. The hiring manager decides, no committee.

The challenge is a sanitized real problem and you can use any AI tools you like. We read what you verified, what you refused to trust and how you knew it worked, and we want to hear about something you killed.

No ghosting at any stage, and every no comes with a reason.

Apply

Tell us about yourself and the hardest problem you went after without being asked.