Three steps: Discovery → Pilot → Rollout
We don't like crash projects. AI solutions get good when you find out early whether a use case actually holds — and when you keep pilot and rollout cleanly separated.
Step 1 — Discovery (1 week)
We start with a 30-minute call, no obligation. If there's a fit on both sides: a structured one-week discovery, ending with a written recommendation document. You decide afterwards whether a pilot follows — no commitment.
Step 2 — Pilot (3-6 weeks, individually priced)
A scoped use case with a clear delivery package. We price each pilot individually based on the discovery outcome, because use cases vary too much to offer meaningful flat fees.
Step 3 — Rollout & maintenance (variable)
If the pilot delivers: handover to your team or long-term maintenance by us. Effort scales with actual load.
Four evanto signatures
In every project we keep four promises — regardless of industry, tooling, or timeframe:
🧪 Dry-run before live-run — Before anything is automatically written, you see the full plan. Only when you're satisfied does it run for real.
✋ What we deliberately don't automate — In every solution we document which decisions stay with humans by design — and why. Honesty as the foundation of trust.
🔄 Swappable language model — Anthropic, OpenAI, Mistral, or local via Ollama — a configuration choice, not a code rebuild. Your model strategy stays free.
🇪🇺 EU hosting, on-premise possible — Data sovereignty in practice: personal-data pseudonymisation, local models for sensitive fields, no cloud sync for sensitive data.
When we say "no"
Not every project is a fit — and we say so early:
- When the AI solution is meant to take the human out of a decision a human should be making.
- When the use case sits in "let's see what AI can do" territory and doesn't address a clear bottleneck.
- When the data foundation is missing and can't be built in the short term.
- When the timeline is unrealistic — we don't ship a demo that breaks during pilot week.
A "no" is the most honest way to head off a later mess.
Toolbox
What we use in production — and why: because we've been running it in production for years, not because it's currently hot in the feed.
- Backend: .NET 10 / C# (Microsoft stack) · Python (for AI pipelines)
- Agent frameworks: Microsoft Agent Framework · MCP (Model Context Protocol)
- Vector stores: Qdrant · Chroma
- Language models: Anthropic Claude · OpenAI GPT · Mistral · local Ollama models
- Infrastructure: Docker · n8n · Directus (headless CMS) · PostgreSQL
- Frontend: Astro · TypeScript · Tailwind CSS
30 minutes — when there's a fit
Let's spend 30 minutes on your specific bottleneck. No obligation, no cost, with concrete next steps at the end.