Features Overview

300 Hour experience - Key Tech

In summer 2021, I interned as a Mechanical Engineer at Key Tech. The firm consults on product development in the Med Tech sphere. As such, I am unable to discuss the name of the client, the device I worked on, or specific challenges in the design. As much detail as I am cleared to release may be found in the relevant project page here.

This is a diagram I made of a typical week’s workflow at Key Tech. We would interpret requests from our client, develop potential solutions, then I would design setups, order parts, and evaluate feasibility. I would either present to the internal team during our team meetings or, if the situation was urgent, spin up the Project Manager and present to client same-day. I deeply enjoyed the time spent in the lab doing things that created new knowledge for the company. I became an expert on applied spectrophotometry, learned how to spec O-rings, and order prototypes while managing supplier relationships. The most delicate balance was between coming up with innovative solutions to tough design problems while relying on proven technologies to avoid adding unnecessary risk to the project. When designing SLA protoypes to meter to microliter precision tolerancing is everything, and I had to critically evaluate my design to see if it was both functionally perfect and possible to manufacture. Even with tolerances of a thousandth of an inch, it was a critical design constraint. I also learned how to work in an office, when to work overtime and when to push a task to the next week, and how to communicate complicated ideas clearly to non-technical personnel. We had a representative from our client visit, and I was able to go out to dinner with them and the team to observe what professional relationship-building looks like in practice. Throughout the experience I enjoyed trainings on important workplace topics like the engineering approval process, what it means to put your signature on a design, cybersecurity, quality assurance, and more.

One of my biggest takeaways from the experience was how feasible adding value actually is. In the global economy it sometimes feels as if all products and labor have been fully commodified and prices are driven so low that startups could not possibly compete with the economies of scale. However, I noticed that there are indeed many opportunities to gain an edge. In a rapid prototyping environment, prices often become secondary to lead time. I asked my PM if a company could manufacture molds for injection molding in 24 hours but at 10x cost and be successful. The answer was an unqualified yes: the actual time value of an unproductive team is a commonly-forgotten danger. When deciding if purchasing specific components to improve test efficiency, I got to value my time at almost a hundred dollars an hour. This fundamentally changed the way I view purchases, and even things such as file conversion software. An hour spent trying to use existing company resources to get files off my iPhone and onto my work PC would cost the client far more than simply buying a specialized flashdrive. Any opportunities to solve specialized needs that many such companies experience may seem like a poor value proposition to an average consumer, but a great change for a large company.

150 hour experience -

In summer 2020, I participated in the Duke Data+ Program. Our client, Fleet Management Limited, tasked our team with using Python in an AWS Sagemaker environment to come up with innovative metrics that could predict costly and deadly incidents. I am unable to detail any results due to IP concerns, but learned a lot about computer and data science that will be discussed through my artifact.

In lieu of providing an actual example graph from our 30+ page report, I have included a nearly-identical graph but with completely different data from the Python library’s documentation. A near-miss was defined as a recorded incident in the ship’s logbook that either was prevented or insignificant enough to the operation of the ship to fall short of incident status. An incident costs at least $10,000 in damages or results in serious injury (or even loss of life). We were able to compare different parameters in different axes to generate heatmaps of different near-miss or incident types as either time series and across ships. The client was particularly interested in time series data, because it was able to validate their gut feelings with hard data. Acting on a hypothesis that a pattern near-misses could be correlated to an incident for a given ship, we applied various numerical techniques to algorithmically correlate the two. One of most successful metrics was determining a danger score as a integrating function across time, attempting to quantify a ship’s propensity to have a near miss. Then, I worked on predicting that number using machine learning toolkits within Amazon Sagemaker.