Summary
Multi-faceted Software Developer with a passion for firmware, electronics, and teaching; seeking employment in senior-level individual contributor roles.
Technical Skills
Professional Experience
Lead efforts to support online field equipment configuration in Ecolab's Industrial IOT devices, enabling customers to remotely manage installed equipment. Re-architected existing software to support new features and reduced CPU usage by more than 50% simultaneously.
Improved stability of Bluetooth LE-enabled networking with deployed devices from Ecolab's IOT gateway. Substantially improved the round-trip time and reliability of data delivery in noisy environments.
Ported mission-critical software to a modern Linux kernel, which improved performance and reliability on aging hardware.
Developed a Yocto build environment for a legacy, unsupported System-on-Module, and created reproducible builds compatible with existing hardware and software. Produced build definitions for Ecolab software which integrate legacy code with modern Linux userspace daemons.
Modified open-source networking stacks to improve BLE advertisement data throughput on Linux. Traced performance bottlenecks to legacy software and demonstrated improvements through replacement of that software with modern alternatives.
Advised the design of high-reliability, long-term scientific data storage on embedded systems.
Shipped bootloaders for new Electronic Control Units in Model Y during a last-minute part swap, in advance of schedule and working across time zones. Eliminated line-down situations in factories and prevented $16M in hardware recalls and service.
Analyzed CANbus performance and identified errors in bit timing on Optimus, improving UDS data delivery reliability by 100% during bus-heavy conditions. Implemented network improvements across all 40+ ECUs on Optimus, and provided a process to safely roll-out network changes to the fleet.
Developed internal tools for bootloader development and firmware deployment in factories, using Python with modern asynchronous IO paradigms, enabling standard laptops to achieve sub 20ms accuracy of protocols. Improved factory flashing performance and reliability by over 50%.
Brought up a complete firmware stack from scratch, starting on pre-production silicon. Created firmware to prove out reliability and silicon-level bugs to inform further silicon development. Shipped 10+ ECUs using this processor in Tesla Model 3, Y, and Cybertruck, reducing lifetime program cost by $10M while improving system performance.
Analyzed motor control performance and LIN traffic to pinpoint firmware bugs where debug access was impossible. Provided in-depth analysis to the controls team which resulted in the solution to a bug which had been under investigation for more than a year.
Became the platform expert on Renesas RL78 and RH850, NXP S32K, Infineon TLE986, and Melexis MCUs. Advised software design and provided insight for debugging hardware-specific issues on these platforms company-wide.
Automated triage of sleep/wake bugs on Mac platforms by writing a decoding framework for Intel Machine Check exceptions. Deployed to the Apple Crash Report pipeline where it handles thousands of panic reports per day.
Enabled early-boot debug information to carry through to Crash Reporter via data export to off-chip storage and retrieval after reboot. Enabled cross-comparison and de-duplication of sleep/wake bugs on all T2 Mac platforms.
Spearheaded conversion of internal tooling from Python 2 to Python 3, while still supporting legacy machine-generated Python 2 code from Intel. Wrote a framework to enable transparent calls between Python 2, Python 3, and Swift 5.0.
Improved Mac EFI builds by creating an automated converter from the EDK II build system to Xcode projects. Improved build times by over 20x and collaborated with the Xcode team to improve build performance for all Xcode projects.
Developed a method for verification of identity from soft biometrics, such as walking gait, heartbeats, and typing dynamics. The resulting architecture is now patented.
Created training and data processing frameworks allowing resource-constrained computing workstations to handle multi-terabyte datasets transparently, without exhausting system resources. Enabled developers to test deep-learning models at their desks. Reduced iteration time of experiments more than 10x.
Patents: US 107682259 B2, US 10768260 B2
Notable Personal Works
Education
Rochester Institute of Technology, Thesis in Adversarial Robustness
Additional coursework in digital design for FPGAs and ASICs in VHDL, as well as real-time and embedded development in C.
Thesis Topic: Adversarial Robustness & Generalization in Image Classification with Deep Neural Networks.
Rochester Institute of Technology
Minor in Mathematics.