AI Agents Built for Real Users
I help product teams turn fragile AI prototypes into agent systems their teams can maintain. We start by scoping the behavior, tools, risks, and quality criteria, then implement and deploy with evals, observability, and documentation your team can use.
Sound Familiar?
- Your prototype works in demos but fails with real inputs, permissions, or edge cases
- The agent loses context, calls the wrong tools, or produces answers you cannot reliably evaluate
- You need production patterns like request throttling, fallbacks, retries, and structured errors
- One agent is trying to do too much, and quality drops as you add more capabilities
- Generic frameworks are no longer enough because your product needs domain-specific behavior
How It Works
Architecture & Design
I analyze your use case and design the agent architecture: single agent or multi-agent, required tools, memory strategy, failure modes, and guardrails.
Build & Integrate
I implement the agents, tool calls, MCP integrations, memory, orchestration, and business rules your product needs.
Harden for Production
I add the boring but necessary pieces: request throttling, model fallbacks, retries, structured errors, input validation, output guardrails, and cost controls.
Evaluate & Ship
I set up evals, deploy with monitoring, and hand over documentation so your team can maintain and extend the system.
What's Included
- Agent architecture design — single agent or multi-agent orchestration
- Tool-calling implementation for APIs, databases, files, and external services
- Memory systems — short-term context and long-term knowledge persistence
- MCP integrations to connect agents with internal tools and services
- Multi-agent orchestration — task decomposition, routing, inter-agent communication
- Production hardening — rate limiting, model fallbacks, error recovery, cost optimization
- Evaluation pipelines to measure and track agent quality over time
- Deployment, monitoring, and documentation for your team
Who Is This For?
- Product teams adding AI agent capabilities to their SaaS or platform
- Startups that need a working agent system, not a 6-month research project
- Companies with a chatbot prototype that needs to become production-ready
- Teams whose single-agent solution has hit quality or capability limits and needs multi-agent architecture
Pricing
Discovery
€3,500
Five-day project to prove feasibility and build a working prototype
- Architecture review and written feasibility report
- Working prototype agent (non-production)
- Tool-calling proof of concept and evaluation recipe
- Clear scope and fixed quote for follow-up work
- Recorded walkthrough and documentation handover
- Fixed price — no hourly surprises
Agent
From €5,500
One production-ready agent for a single use case, deployed to your cloud
- 1 agent with tool-calling, memory, and business logic
- Integration with your existing stack
- Production hardening (error handling, retries, fallbacks)
- Evaluation pipeline with regression tests
- Deployment to your cloud with CI/CD
- 14 days post-launch support
Need a larger project?
Multi-agent systems (4–6 weeks) and platform-scale rollouts (8–12 weeks) start at €22,000 and are scoped during the Discovery sprint, which becomes the first week of every larger build. The same applies to framework migrations: effort depends heavily on the current codebase, so pricing comes after scoping.
Common Questions
Start with a single agent unless there is a clear reason not to. Multi-agent architecture makes sense when quality drops as tools are added, tasks require different areas of expertise, or work needs to happen in parallel. If the right shape is unclear, the 5-day Discovery tier (€3,500 fixed) delivers a prototype, a feasibility report, and a firm quote for the follow-up work.
Primarily TypeScript and Python. I work with any LLM provider (OpenAI, Anthropic, Google, open-source) and integrate with your existing stack — whatever framework, database, or cloud you're already on. Common agent frameworks: Claude Agent SDK, LangGraph, OpenAI Agents SDK, CrewAI, or hand-rolled.
Discovery is a fixed 1-week project. Agent builds usually take 2–3 weeks. Larger projects, such as multi-agent systems (4–6 weeks) or platform-scale rollouts (8–12 weeks), are scoped during Discovery so the final timeline and price are grounded in the actual code and product requirements. Rush start is available at +25% when the project needs to begin within 2 weeks.
Yes. I deploy to your cloud (AWS, GCP, Vercel, etc.) with CI/CD, set up monitoring and alerting, and provide documentation for your team to maintain.
Every build includes documentation and a post-launch support window (14 or 30 days depending on scope). Ongoing support is available for prompt tuning, eval monitoring, model upgrades, and feature additions, and is scoped after the build phase instead of sold as a generic package. Typical support plans run 3–12 months.
Yes. The Agent Code Audit is a fixed 3–5 day project (€3,500) that covers architecture, eval coverage, prompt-injection risk, and cost. Framework migrations (LangChain → LangGraph, v0 → Claude Agent SDK, hand-rolled → Mastra, etc.) are scoped in Discovery first because effort depends heavily on the current codebase.
Yes. That's what Discovery is for: five days, €3,500 fixed, a working prototype, a written feasibility report, and a firm quote for production work. It gives you a way to reduce risk before committing to a larger Agent or System project.
Let's Talk Through the Project
Tell me what you're building, where the AI is getting stuck, and what needs to work in production.