Most AI features demo well and fail quietly. The difference is almost never the model: it is the retrieval, the evals, the guardrails and the monitoring around it. I build the parts that were never intelligent, because they are what make the intelligent part safe to use.
Fixed scope and a fixed price, both agreed in writing before anything starts. If your project sits between two of these, it gets its own number.
from€4 000
1–2 weeks
Best when a single AI feature is standing between you and a release.
Includes
Recommended
from€10 500
3 weeks
Best when the value is locked inside documents nobody can search.
Includes
from€21 000
6 weeks
Best when the AI is the product and it has to hold up.
Includes
Still deciding, or is the scope still moving? A paid scoping call (€400) puts the plan and a fixed number in writing, and is credited against the project if we go ahead. The thirty-minute intro call stays free either way.
The same four steps every time, so you always know where it is.
A few lines is enough. I read every message myself and reply within one business day.
Free, and no pitch. We work out what done means here, and whether I am the right person at all.
If the scope is clear after that call, the plan and the fixed price follow it at no charge. If it is still moving, the paid scoping call settles it first and comes off the project.
A working demo every week, not a status report. You find out in week two if something is wrong, not in week eight.
Whichever fits: OpenAI, Claude, Gemini, or open-weights models running on your own infrastructure when the data cannot leave. It is a constraints question, not a preference.
Yes. I have deployed on Azure, AWS, GCP and on-premise Kubernetes, including self-hosted open-weights models where compliance required it.
By grounding answers in retrieved sources, showing citations, and running an eval set that catches regressions before your users do. Anything the system cannot answer should say so rather than guess.
Yes. Most of this work is joining something that already exists rather than starting a new repository.
All of it, in your repository, with the infrastructure defined as code.
A function with an LLM call inside is not an agent, and calling it one is not a harmless simplification. I built the same expense-audit task as a function, a workflow, and an agent, ran 150 graded runs, and measured what the word actually costs, then scaled the data until the function's only move stopped existing and the agent was the last architecture standing.
Field notes from building Smithproof, a multi-tenant control plane for AI agents, and why detection without response and policy without enforcement govern nothing.
Field notes from OrbitSpec, a clarification engine that turns vague requests into typed, auditable Context Bundles before any model runs, and why reliability starts before the answer.
Tell me what it has to answer and which sources it can read. I'll send back how I would build the retrieval, how I would measure the answers, and what a request would cost.
I read every message myself and reply within one business day. No sales pitch, just a straight conversation about your build.
Prefer to hire through a marketplace, with escrow and platform reviews? This service on Fiverr
Last updated