The trap
Every product company is under pressure to "add AI". Most of it is malinvestment: a chat box nobody asked for, demoed once, then left to cost money in API bills and maintenance. I start with a different question: what do customers already ask for that AI makes possible or cheaper?
What I build
LLM features inside your product: summaries, drafting, classification, search, extraction, and agents that complete multi-step tasks for your users.
Human-in-the-loop flows. The AI proposes, a person approves. This applies wherever a wrong answer costs money, trust or compliance.
Production basics: evaluations against real data, cost and latency limits, fallbacks when a model fails, logging, and prompt and model versioning.
Model choice that fits: OpenAI, Claude, open models, or a mix. Picked for cost and quality.
How I work with your team
I work inside your codebase, with your engineers. I build the first feature with them, set up the patterns and guardrails, and leave your team able to ship the next one without me. Your team sees how everything works and doesn't need me to keep it running.
Why me
At sennder, a fast-moving logistics company with a complex org, I was a senior software engineer and then an engineering manager. I know what it takes to ship inside a real company with real constraints and other teams depending on you. I also build AI systems for myself every day: agentic tools with fixed steps and AI steps, and a team of bots that runs product work from planning to release with a human supervising. If your stack uses something I haven't worked with, I learn it.
FAQ
Which AI feature should we build first?
The one tied to something customers already ask for or already pay for. That's usually a slow, manual step in their workflow. I help you find it before anyone writes code.
How do you stop the AI from making costly mistakes?
With evaluations on real data before launch, strict limits on what the model can do, and human approval on actions that matter.
Will this blow up our API costs?
Not if it's designed for cost from day one. That means the right model per task, caching, limits, and monitoring of cost per feature.
How long does it take to ship a first AI feature?
It depends on the feature and your codebase. I'll give you a real estimate once I've seen the code.
Do you replace our engineers?
No. I work with them and leave them able to run and extend what we build.