I appreciate systems like actor-critic that divide a continuous action space problem into parts that each may require traditional ML, search ML, or control system algorithms. Ultimately, I want the state of the art to incorporate cognitive heuristics and interpretable representation patterns.


I have five years of engineering work experience, divided between software, machine learning, and mechanical.

Robotic design brought me to automation, for manufacturing minimally invasive surgery devices. Then I transferred to UCSD for Machine Learning and Neural Computation, where I’ve explored so many of my interests with terrific resources and faculty.

I feel like software engineering enables the same abstract creative thinking I treasured in mechanical design, while providing a limitless toolkit for developing intuition and scaling effectiveness.

At Endura Technologies our product involved an audio application for deep learning. The algorithm was ultimately visual based, and a pipeline filtered and re-represented audio training snippets for it. At UCSD, I took Neural Signal Processing, which improved my toolkit and understanding in this area. After that, I supported an applications engineer with testing and process automation.

Short list of highlights:

  • Coded readable, interfaced, faster runtime testing routines that used to take weekends to run and half a week’s worth of hours to be manually altered per test.

  • Automated several steps of long and error prone coding by generating java code from an excel spreadsheet of register functions using python.

  • Designed, manufactured, and qualified more than seventy End-of-Arm tooling for the polymer injection moulding process using stock and custom machined grippers and linkages, being assigned the most challenging molds that needed creative solutions for retrieval and sensing. Each tool saves thousands of dollars per run from mold damage and variance, as well as hundreds of man hours otherwise spent separating part and sprue.

  • Collaborated with executives and the automation department on larger projects, aiding in design choices and product development as well as user experience, troubleshooting and qualification.

  • A heavy tool kept misaligning with the mold, damaging ejection pins and not retrieving parts consistently. Designed and constructed double spring loaded alignment pin retrofits that not only saved $10,000 per month, but had the most satisfying tactile feedback.

Story about my last project at Applied Medical:

A cutting edge metal injection insert moulding process for a new suturing device was costing $100,000/month in variance from pin misalignment, as well as being a super slow and awkward task for production.
My guess was that the magnetic pen being used was causing some pins to polarize and repel from their channel. After ordering some fishing line variants and a suction pen intended for microchips, I 3D printed a perfectly sized grid with rods and imprints for the fishing line to fold into, and designed stainless steel suction pen tips. Pouring in the two-pronged pins, they would face the same direction hanging between the lines, with the aberrant ones falling through the grate and pooling to the corner for recycling. The best of my pen tips then created a vortex around the cap of the pins so that they stayed set at a predictable angle rather than collapsing to a different orientation or changing angle when pressed against the insert mold. It also had a notch to tune the operator in with the insert mold channel.
When implemented into production, it completely solved the variance issue, and sped up the shop orders significantly. I had a working prototype ready soon after the ordered components arrived, thanks to effective prototyping and vertical integration.