Nathaniel Mapaye

EmbNode

Embedded telemetry firmware for ESP32 and STM32 boards, written in C++17 on FreeRTOS. Sampling, batching, and sending each run as their own task, and the packet format, OTA update logic, and power accounting are tested on a laptop before any hardware is involved.

Problem

A battery-powered sensor on an unreliable network has to keep sampling on schedule when sends fail, and an OTA update cannot leave the device in an unknown state.

System design

Sampler, aggregator, and communications tasks hand data off through FreeRTOS queues. Communications tries MQTT first and falls back to HTTP.

Evidence

Host tests feed in truncated frames, bad CRCs, and OTA backends that fail or commit after the deadline, with a fake clock so timing is repeatable.

Outcome

The packet and OTA logic pass those tests. DMA sampling, delivery retries, and battery measurements on a real board are still to do, and a full queue currently drops samples without counting them.

View EmbNode sourceRead the telemetry design

SysLib

A header-only C++20 library with two queues. The SPSC ring is lock-free and hands values from one producer to one consumer with acquire/release atomics. The MPMC queue is a deque behind a mutex and condition variable, so waiting consumers sleep instead of spinning.

• Ring capacity rounds up to a power of two, so index wraparound is a bit mask
• If moving a value out throws, the queue wakes the next consumer so the item is not stranded
• Tests run four producers and four consumers and check that every value arrives exactly once
• Optional ThreadSanitizer build

View SysLib sourceRead the concurrency design

GitOps

A Kubernetes operator in Go that watches a canary release. It reads p95 latency from Prometheus, runs Gatekeeper policy checks, and rolls the canary back when latency regresses.

• Rollback finished in under 30 seconds after a p95 regression, measured in a test cluster

View GitOps source

AURORA

An iPhone and iPad app for logging caffeine, checking daily intake, and running a 60-second reaction test for alertness. Sleep data comes in through optional read-only Apple Health access, and everything stays on the device.

Problem

I wanted to see how late-day caffeine affected my alertness without sending caffeine and sleep data to a server.

System design

React Native with a Swift bridge to HealthKit. Sleep samples are cleaned up (reversed intervals and duplicates dropped) before the rest of the app reads them.

Evidence

Tests cover Health access granted, denied, unavailable, and empty, plus malformed and duplicate sleep samples.

Outcome

The app works fully with Health access denied. I use the reaction test before long drives, and my friends use it to compete over who is most alert.

Read AURORA case studyView AURORA demo