Eval suites
We build domain-specific test sets from real examples — auto-graded, with a clear pass criterion per use case.
Prompt Engineering
We treat prompts like software: with an eval suite, a diff workflow and CI. So your model delivers consistent quality, not situational sparks.
Iteration
Three versions of the same prompt, three eval runs, three measurable steps. This is what our iterations look like.
Capabilities
Prompt engineering is more than word choice. It is test discipline, architecture and knowledge work combined.
We build domain-specific test sets from real examples — auto-graded, with a clear pass criterion per use case.
Roles, output schemas, tool specs and few-shot strategy — cleanly separated, versioned, in the repo.
We hand over the craft — playbooks, internal training and review sessions — so your team keeps prompts evolving.
Diff example
Every change lives in a pull request, runs the eval suite, and only lands in main when the numbers move.
<task>- Summarise the ticket.</task>- Answer:
<role>support_lead</role>+ <task>Summarise the ticket in 3 sentences.</task>+ <output_schema>+ { summary: string, urgency: "low"|"med"|"high" }+ </output_schema>
Eval results
Eval suite with 240 real tickets from a mid-sized mobility provider service desk — same model, only prompt work.
Pass rate
42%→93%+51pp
Schema adherence
55%→99%+44pp
Hallucinations
18%→3%-15pp
Tone of voice fit
60%→91%+31pp
Cost per 1k calls
62 €→31 €-31 €
By the numbers
Engineers, designers, and strategists working as one practice from our Hamburg HQ.
Companies across consumer, healthcare, and B2B trust us with their digital products. Long-term partnerships are the default.
Repeat engagements and references our buyers actually call. Trust compounds when delivery does.
Custom mobile and web products shipped from concept to maintenance — owned end-to-end by our team.
Strategy, design, and engineering all live at our Hamburg HQ. One team, one project lead, accountable to you from kickoff to launch.
Helping companies ship digital products — and growing alongside the teams we work with.
Next steps
Book a 30-minute discovery call. We'll review your goals, surface unknowns, and outline how we would run the engagement.