ai agent development
AI agents that
do the work.
I design and build custom AI agents — copilots, assistants, role-play coaches, and autonomous workflows on Claude and GPT — with the guardrails, evaluation, and cost control that keep them working past the demo. A flat €25 / $25 per hour, and there's a live one below you can try right now.
Within 24 hours
You message me and I reply myself — not a sales queue, not an account manager.
A live agent, shipped
Not a slide deck — a real LLM agent in production you can open and talk to in the next minute.
Built to survive users
Guardrails, evaluation, and cost caps designed in — not a demo that falls apart in week two.
€25 / $25 per hour
One flat hourly rate, your choice of currency — for a prototype or a production agent.
01 — live demo
Talk to an agent I built.
The fastest way to judge AI agent work is to use one. This is a real product, running in production — no login, no signup.
Skillforge — AI Communication Coach
A conversational AI coach that runs live LLM role-play: you practise a tough conversation, the agent plays the other side and gives feedback. It ships with the unglamorous production parts that actually matter — a per-browser usage quota, a global cost cap so a viral day can't blow the budget, no-login access, and streak tracking. Same engineering I'd bring to your agent.
02 — what I build
Agents, not just API calls.
Anyone can wire a prompt to a chat box. The work is the part that makes it trustworthy: retrieval, tools, evaluation, guardrails, and a cost model that doesn't surprise you. That's what I build.
Product copilots & assistants
An assistant embedded in your product or workflow that actually knows your domain — answering, drafting, and acting on your data instead of guessing from thin air.
- In-app chat copilots
- Drafting & summarization
- Support & onboarding assistants
- Domain-grounded answers
- Streaming, citations & memory
- Claude, GPT & open models
Role-play & coaching agents
Agents that hold a character and a goal across a conversation — practice partners, tutors, interviewers, and coaches — like the live Skillforge demo above.
- Persona & scenario design
- Live multi-turn role-play
- Feedback & scoring
- Session state & progress
- No-login or account-based
- Abuse & safety guardrails
RAG & retrieval pipelines
Give the agent your documents, database, or knowledge base so its answers are grounded in your reality — with the retrieval, chunking, and evaluation to keep it accurate.
- Document & data ingestion
- Embeddings & vector search
- Grounded, cited answers
- Freshness & re-indexing
- Retrieval quality evals
- Hallucination reduction
Autonomous workflows & tools
Agents that don't just talk — they call tools, hit APIs, and complete multi-step tasks, with the right guardrails and a human in the loop where it counts.
- Tool & function calling
- API & system integrations
- Multi-step task automation
- Human-in-the-loop approvals
- Evals & quality monitoring
- Cost & rate-limit control
03 — how it works
From prototype to production.
A low-risk path that proves the agent is useful before you invest in hardening it. A flat €25 / $25 per hour throughout.
Define the job
What should the agent do, for whom, grounded in what data? We pin down the task and how we'll know it works.
Working prototype
A real, usable agent fast — so we test the idea against actual inputs instead of arguing about it.
Evals & guardrails
Evaluation suite, safety guardrails, cost caps, and monitoring — the difference between a demo and a product.
Ship & maintain
Deploy into your product or infrastructure, then improve it on a retainer or hand it over. It's yours.
04 — questions
Frequently asked.
Which models do you build on?
Claude and GPT primarily, plus open models where they fit. I'm not tied to one vendor — I pick per task and can route between models for quality and cost.
Won't it hallucinate or go off the rails?
That's the engineering. Retrieval grounds answers in your data, guardrails constrain behavior, and an eval suite catches regressions — the same approach that keeps the live Skillforge agent stable.
Can it use my own data?
Yes — that's usually the point. I build RAG pipelines over your documents, database, or knowledge base so the agent answers from your reality, not generic training data.
How do you keep API costs from exploding?
Cost caps, per-user quotas, caching, and model routing, designed in from the start. The Skillforge demo runs a hard global cost ceiling so a viral day can't blow the budget.
How much does it cost?
A flat €25 / $25 per hour — your choice of euros or dollars — for a prototype or a production agent. You get a quote up front so you know the scope before I start.
Is it one-off or ongoing?
Either. Build a prototype and stop, ship a production agent, or keep me on a retainer to evaluate, tune, and extend it as your needs grow.
get a quote
Describe the agent you need.
A few sentences is plenty — what it should do, who uses it, and what data it should draw on. I read every message myself and reply within 24 hours with an honest take and a quote at €25 / $25 per hour.
prefer email?
admin@agicad.com