I help product teams turn flaky systems into reliable ones — stabilizing embedded firmware, hardening cloud backends, and adding AI agents that actually work in front of real users.
Production Systems Consultant
Firmware in the field, APIs under load, AI agents in front of real users — the same pattern keeps repeating. The demo is great. Production is fragile. The launch slips.
Devices reset, watchdogs fire, and you can't reproduce it on the bench.
Heap shrinks, fragmentation grows, devices die after 24–72 hours of uptime.
Partial flashes, signature mismatches, rollback loops — and field recalls.
Drops, reconnect storms, broker disconnects, and silent data loss in the cloud.
APIs time out, queues back up, dashboards lag — and nobody can tell if it's the device or the cloud.
Demos impress. Real users hit edge cases, the agent hallucinates, costs spike, and trust erodes.
"Almost ready" has been true for three months. Every sprint adds bugs faster than it removes them.
I embed with your team, find the real root cause, and leave you with a system that ships — plus the documentation and tests to keep it that way.
Devices stay up for weeks, not hours. Crashes, watchdog resets, and field returns drop to a level your support team can actually handle.
Stable connectivity, predictable timing, and a memory profile that survives 24/7 — instead of falling apart after the first soak test.
Safe OTA, secure boot, and a real recovery path so updates become routine instead of the most terrifying day of the quarter.
A cleaner architecture, written tests, and clear docs — so your engineers can keep shipping after I'm gone.
APIs and AWS IoT pipelines that scale with your fleet, recover from outages on their own, and don't surprise you with the bill.
Focused agents with clear tools, guardrails, and evals — so the answers are right, the costs are bounded, and your team isn't babysitting the model.
A few examples of stabilization work for IoT and hardware teams.
Devices were rebooting daily in the field. Heap shrank over time and the watchdog tripped under MQTT load.
Profiled heap and stack usage, fixed a handful of leaks, restructured FreeRTOS task priorities, and replaced ad-hoc buffers with bounded queues.
~80% fewer crashes and 7+ days of continuous uptime under full load. Launch unblocked.
Roughly 1 in 50 devices bricked during OTA. No rollback path, no integrity checks, support tickets piling up.
Implemented dual-bank OTA with signed images, atomic boot-flag handling, and automatic rollback on first-boot failure.
Zero bricked units across the next rollout. Field updates became a routine, not a risk.
When Wi-Fi flapped, hundreds of ESP32 devices reconnected at once and overwhelmed the MQTT broker, causing data loss.
Added jittered exponential backoff, persistent sessions, QoS tuning, and a local message buffer to survive outages gracefully.
No more reconnect storms and zero telemetry loss during network events. Backend cost dropped too.
Fixed scope. Fixed timeline. A specific outcome by the end — not a stack of hours.
A fast, deep review of your firmware, backend, or AI agent — so you know what will break in production before your customers do.
I embed with your team and turn an unstable system into one that runs reliably under real load — whether it's firmware, a backend, or an AI agent.
Wire your device to a real backend — and, if you want, an AI agent on top — so customers see one product, not a stack of half-finished pieces.
I've spent 10+ years shipping embedded products: ESP32, STM32, Nordic nRF52, FreeRTOS and bare-metal firmware, BLE, Wi-Fi, MQTT, and OTA pipelines on AWS IoT.
Beyond firmware, I also build the web services (Lambda, API Gateway, DynamoDB) and AI agents (LLM tool-use, retrieval, automation) that sit on top of device data — so the whole stack, from MCU to cloud to user, actually works together.
I focus on the messy part — the bugs that only appear in the field, the OTA flow nobody trusts, the memory leak that takes a week to surface. I work directly with your engineers, fix the real issues, and document what changed so your team owns the result.
Book a free 45-minute call. Bring the problem — device, backend, agent, all of the above. I'll tell you straight whether I can help, and what it would actually take.
Tell me a bit about your project and the issue you're seeing. I read every message and reply personally, usually within 1 business day.