Services
Four fixed-scope engagements. One senior person owns the whole thing, from the first question to the launch.
2–3 weeks · from €6k
AI Product Sprint
From idea to an evidence-based go or no-go.
- Market and competitor research with dated sources
- Unit-economics model: cost per user, cost per request, margin at scale
- Architecture and model shortlist, with a small bake-off on your real inputs
- A working prototype of the core loop
Proof: Clearcut, Glyco and Photco research; the HireMike voice-stack decision.
6–10 weeks · from €25k
Idea to Launch
A production MVP that can take payments on day one.
- Auth, billing, analytics, status page and cost alerts
- Built in vertical slices; you verify each one before the next starts
- CI, docs and a handover, so your team or the next engineer can continue
- You own the code, the cloud accounts and the data
Proof: Clearcut went from first commit to paid launch in 6 days.
1–2 weeks · from €4k
AI Reliability & Cost Review
For teams with an AI feature that is flaky or expensive.
- An eval harness on your real traffic, with quality, latency and cost per request
- A model bake-off, including cheaper and open-weight options
- Fallbacks, guardrails and caching where they pay off
- A written report with the changes ranked by impact
Proof: Agent eval harnesses, Whisper hallucination gating, rule-oracle bake-offs.
2–4 weeks · from €8k
Agentic Engineering Setup
Coding agents working inside your team, safely.
- Ticket to reproduce-first test to draft PR, running in your CI
- Review agents and PR-comment responders
- Claude Code or Codex harnesses with skills, hooks and permissions
- A buildability agent that checks designs against your codebase
Proof: Jira-to-PR agents and a Figma buildability agent in production use.
- How much does it cost to build an AI MVP?
- A production MVP with auth, billing and analytics starts at €25k and usually takes 6–10 weeks. If the idea isn’t validated yet, start with a 2–3 week AI Product Sprint from €6k. It ends with a prototype, a cost model and a go or no-go.
- Do you work with non-technical founders?
- Yes. You get a written brief, a decision log and a working slice to try every week, in plain language.
- Which models and clouds do you use?
- Whatever the measurements favour. I have shipped on Claude, GPT, Gemini, Amazon Nova and open-weight models, on AWS (Bedrock, Lambda, Step Functions), Cloudflare Workers and serverless GPUs. I pick models with an evaluation and a cost per request, never by habit.
- Who owns the code?
- You do: the repository, the cloud accounts and the data, from the first day.
- Where are you based?
- Sarajevo, Bosnia and Herzegovina (CET). I work remotely with teams in the EU and the US.