Not demos. Systems that take real phone calls, move real money and dispatch real drivers at 2am while I'm the only engineer awake.
Self-taught. Writing code since 17. Five years in production. When a problem catches me, I don't let go until I understand it better than the people who reported it. Then I build the thing that makes it disappear.
Close to the problem. I sit with the ops team, listen to the support calls, read every complaint, watch how people actually use the product. Then I go back, build the fix in days, ship it, and live with it in production.
The gap between "what the customer said" and "what the customer needs" is where I like to work.
First engineer at an AI-first logistics startup. Took it from an empty repo to 65k+ active users. Ran engineering solo for over a year and a half.
- A conversational AI that runs the business. Customers message an assistant, it books pickups, drops and shopping end to end. Agent orchestration on LangGraph, with custom components wherever off-the-shelf tools broke down.
- A voice agent that takes orders over the phone. Support calls were drowning the team. Now an AI answers, takes the order and helps the customer.
- A privacy layer between customers and drivers. Complaints showed drivers calling customers' personal numbers. Built number masking so neither side sees the other's real contact.
- The dispatch brain. Driver assignment, live tracking and a driver incentive system.
- Replaced paid SaaS with in-house systems. Accounting reports, ops workflows and call automation, all built in-house, cutting external tool costs.
Before AI, I built the hard plumbing: a video streaming platform, a delivery logistics app, and a file compression and sharing platform that was later acquired. Then shipped frontends for multiple client products at a software agency.
- Field first, code second. Talk to the people closest to the pain before touching the editor.
- Prototype in days, harden in production. Get something real in front of users fast, then make it bulletproof.
- Own the outcome, not the ticket. If the fix needs a backend change, an ops process change and a conversation with the customer, I do all three.
- Make it make sense to non-engineers. Ops, finance and founders should understand what got built and why.
- Stay calm when production isn't. Most of what I know, I learned debugging live systems with users waiting.
- Making LLM agents behave the same on the 10,000th conversation as on the first: guardrails, fallbacks, evals
- Voice AI latency, where 300ms decides whether it feels human
- Agentic engineering: using AI coding agents with real architecture, rules and review, not vibes
LangGraph · LLM tool calling · voice and telephony · CI/CD
If you have a messy, real-world problem and want someone who'll go to where it lives and come back with it solved, reach out.