Embedded Portfolio
Four production firmware builds: autonomous greenhouse control, maritime edge compression, EV fleet gateways, and Modbus energy bridges. Each card represents a system where timing is a requirement and safety is the priority. Flip them to see the firmware levers behind the silicon.
Selected Work
Real-Time Control
Sub-50ms climate control loops across 12 independent greenhouse zones using STM32H7 and FreeRTOS.
Contribution: Designed a custom PCB and firmware with deterministic CAN bus backbone for multi-zone synchronization.
Impact: Eliminated timing jitter and reduced crop stress incidents to zero in the first growing season.
Autonomous Climate Controller
The Hard Part
Greenhouse environments shift rapidly. A jittery control loop means vents open late or heaters overshoot, directly impacting crop yield. Standard PLCs often lack the sub-50ms resolution needed for multi-zone sync.
The Approach
We moved the logic to an STM32H7 running FreeRTOS with strict task priorities. The CAN bus backbone ensures every zone receives its control frame within a guaranteed 5ms window. High-frequency PID loops run in dedicated hardware interrupts to ensure zero-lag response to thermal spikes.
Edge Telemetry
Satellite-optimized telemetry ingest on ARM Cortex-M7 for 340+ vessels.
Contribution: Built a custom delta-compression engine and NMEA 2000 sensor ingest layer with satellite-aware buffer management.
Impact: Reduced satellite bandwidth consumption by 68% while maintaining 100% data integrity during outages.
Maritime Edge Data Compressor
The Hard Part
Satellite data costs 1000x more than cellular. Shipping raw NMEA 2000 streams is economically impossible, but operations needs the data to detect engine wear before a failure.
The Approach
We implemented a bit-level delta-compression algorithm on the Cortex-M7. The firmware analyzes sensor variance and only transmits significant changes. During satellite outages, data is buffered to local flash with a priority-aware eviction policy that preserves the most critical health signals.
Smart Gateways
ESP32-C6 based charging controller handling OCPP 2.0.1 communication and contactor safety interlocks.
Contribution: Implemented the OCPP 2.0.1 stack over WebSocket with hardware-level safety state machines.
Impact: Enabled dynamic load balancing for 80+ vehicles with sub-second response to grid pricing signals.
Smart EV Charging Gateway
The Hard Part
Software fails, but 400V contactors must fail safely. If the WebSocket drops or the MCU hangs, the charging current must be cut or maintained according to a strict safety policy.
The Approach
We built a hardware-first safety state machine. A dedicated watchdog timer monitors the OCPP stack; if it misses a beat, the hardware interlocks drop the contactors into a safe state. The ESP32-C6 handles the high-level TLS/WebSocket work while the low-level logic runs in a separate, isolated task.
Protocol Bridges
ESP32-S3 bridge board translating Modbus RTU/TCP to real-time InfluxDB telemetry.
Contribution: Designed a custom Modbus bridge with ESP32-S3 that captures 1-second resolution energy data from legacy BMS systems.
Impact: Provided per-floor energy visibility and enabled anomaly detection that caught a major HVAC failure.
Modbus-to-Cloud Energy Bridge
The Hard Part
Legacy building management systems (BMS) are silos. They speak Modbus RTU over RS-485 at low baud rates, making it difficult to extract high-frequency data for modern AI models.
The Approach
We built an ESP32-S3 bridge that acts as a Modbus Master. It polls legacy meters at 1-second intervals, aggregates the results, and flushes them to a local buffer before pushing to InfluxDB. The firmware handles protocol timing sensitivities and electrical noise common in industrial electrical cabinets.
Firmware Diagnostics
Waiting for system reveal
Real-Time Execution Notes
When the character triggers a card lever, the embedded system diagnostics reveal is activated, surfacing the design decisions behind deterministic edge control.
In embedded systems, correctness is relative to time. A correct decision delivered late is a failure.
Avoid heap fragmentation in long-running firmware. Bounded buffers and static memory ensure predictable behavior.
If a field engineer can't explain a control decision, the firmware isn't finished. Built-in telemetry is mandatory.
The edge must remain safe when the cloud fails. Design local fallback modes that preserve core safety loops.
Development Bench
Optimizing WebSocket-based protocol performance on ESP32-C6 for large-scale fleet charging.
Researching NMEA 2000 delta-compression patterns to minimize satellite telemetry costs.
Designing CAN bus synchronization protocols for sub-1ms alignment across multiple microcontrollers.
Personal Philosophy