AI engineering, from prototype to production.
I build agents, assistants, and AI features with the APIs, permissions, evaluation, and observability that make them dependable in daily use.
| Good for | Turning a use case or prototype into a live system |
|---|---|
| Engagement | Scoped project or $200 / hour |
| First step | A short brief by email. Free. |
What the work involves
AI engineering turns model capability into software people can rely on: choosing the right approach, connecting tools and data, evaluating behavior, and building the application around it.
- Custom agents for operations, customer support, and internal engineering workflows.
- Voice and chat assistants connected to your business rules and tools.
- Backend APIs, interfaces, and tool calling for actions inside existing systems.
- Evaluation suites, error handling, monitoring, and human approval paths.
When it helps
You have a defined use case or an early prototype, and need an engineer to make it reliable with your real systems, data, and users.
A typical project
An engineering team wants help investigating service issues. The assistant retrieves the relevant runbooks, queries approved telemetry tools, and presents findings with the evidence behind them. The implementation includes read-only permissions, test scenarios, and an explicit way to flag uncertainty.
What you get
| Deliverable | What it includes |
|---|---|
| A scoped implementation | Application code and the agreed integrations, with the agent’s supported tasks and limits documented. |
| An evaluation harness | Representative scenarios and agreed measures for answer quality, task completion, latency, and cost. |
| An operable handover | Configuration, operating notes, and a rollout plan covering monitoring and ongoing ownership. |
Questions
Can you improve a prototype we already have?
Yes. I review current behavior, integrations, evaluation coverage, and operating requirements, then scope what’s needed for a pilot or a full rollout.
Do we need to train our own model?
Usually not. I first assess whether existing models can meet the task. Most business workflows need better integration, relevant context, and rigorous evaluation well before they need custom training.