Forward Deployed Engineer (m/f/d)

The European trades are one of the least digitised industries left. Software
penetration sits under 15%, against roughly 70% in the US. The labour
productivity gap runs about 30%. Millions of skilled people spend their evenings
on paperwork instead of the craft they trained for — in the industry that has to
deliver the energy transition.

There is almost no incumbent software to displace. That is unusual, and it won't
last.

smalt replaces the back office entirely rather than selling Meister another tool
to manage. In the last six months that went from a partner app and a dashboard
to around thirty AI agents in production - pricing real jobs, qualifying real
leads, running the operations of businesses whose payroll depends on it working.

These days building thirty agents is easy. Knowing which thirty is what's challenging.

That only comes from being in the Betrieb and talking to the Handwerker.

That's what this role is for.

What you'll do

You'll sit in Handwerksbetrieben - at the Meister's desk, in the van, on site, next to our team -
and watch how the work actually happens. Then you'll come back and build the
thing that fixes it, often multiple times on the same day.

- Spend real time with partners. Not a quarterly customer call. You'll know
  our Meister by name, understand their P&L, and notice the thing they've
  stopped complaining about because they assume it can't be fixed.
- Build across the stack. Python/FastAPI services, Vue interfaces, Flutter
  screens, agent loops, data pipelines - whatever the problem needs.
- Own agents in production. Not demos. Agents that price real jobs, answer
  real customers, and cost real money when they're wrong. Reliability,
  evaluation, structured output, graceful failure.
- Decide what to build. You'll write your own acceptance criteria. Nobody
  here will hand you a spec, and waiting for one is how to fail in this
  role.
- Measure whether it worked. Hours saved per partner, offers that go out
  untouched, leads answered before a competitor picks up. If the number didn't
  move, it didn't work.

What that looks like in practice

In July one of our Meister got an early build of the partner app, ahead of the
general rollout. He tested it between jobs and sent screenshots: from a calendar
appointment he could no longer jump into the project to pull the customer's
phone number. He had a reply in two minutes. Within the hour he'd also asked for
a navigation button - tap the address, go straight to maps.

Neither was news. Both were already being worked on before he hit send.

We aim to be fast, and close enough to the work that we've usually seen the problem before the person living it has to tell us.

You might be a great fit if you

- Have owned something end to end - architecture through delivery through the
  unglamorous iteration afterwards. We don't count years; we look at what you
  carried.
- Get obsessed with a problem rather than a stack. The tools change every few
  months. The curiosity compounds.
- Would rather ship something that changes a tradesperson's week than win an
  argument about abstractions.
- Actually like talking to customers, and have changed your mind about a feature
  because of something a user said.
- Are comfortable being the least knowledgeable person in a room full of
  electricians, and ask the question anyway.
- Can hold "ship it today" and "this needs to survive two years" in your head at
  the same time, and know which one applies.

Our stack, so you know what you're walking into:

- Backend: Python, FastAPI, async SQLAlchemy, PostgreSQL (with pgvector), Celery, Redis
- Frontend: Vue 3, TypeScript, Vite, PrimeVue
- Mobile: Flutter
- AI: Anthropic, Open source and Google models, vector search over partner and catalog data
- Infra: Google Cloud Run, Cloud SQL, Sentry

We care that you're strong somewhere and unafraid everywhere else. If you've
never written Vue but you've shipped production systems and you learn fast, that
is fine - several of us hadn't either.

What we do look for:

- Production experience with LLM-powered features, and the scars that come with it: evaluation, structured output parsing, context management, error recovery
- A track record of shipping things people actually used
- High agency. You'll get context and constraints, not instructions.
- Raw intelligence paired with a humble mindset

German is helpful and you'll pick some up either way - our partners speak it all day. It isn't a requirement but definitely a nice to have. The the team works in English.

- Top-of-market compensation and a meaningful VSOP package
- Direct access to real users and short feedback loops
- Real ownership, and direct proximity to the founders - and to General
  Catalyst, noa, Owl Ventures and Heartfelt
- A team of five engineers who have shipped more in six months than most teams
  do in two years, and who set the pace themselves