Hands-on oversight of the R&D associated with software development involving real-time image analysis, and doing it efficiently on limited hardware in service of understanding what a human is doing while exercising (or getting set up to exercise). Lots of mathematical and algorithmic puzzles to solve. I have been using Rust and Dart for this.

Khaylo has an iOS app that, when you strap the phone to your arm, recognizes what exercises you are doing, counts reps, stores the data in the cloud and allows you to interact with friends about your fitness activities. I have been helping out with the core sensor data processing and machine learning portions of the app, server analytics (Scala/Akka/Spray talking to Parse.com, for now), and some utility software for the data scientist on the team. I learned Swift and began writing modules in it, and helped the rest of the team switch over from Objective C.Khaylo had been working on the problem for 18 months before I joined them, but things were not going well. The CTO, with whom I had worked previously, fired two developers and hired me. It took over a year to get things working, and then we made several rounds of improvements.

Worked on three JVM-based servers with REST APIs: one for virtualization management (the system Coraid acquired from Yunteq) and two storage management servers. I switched the team from Java to Scala on the first server, and the latter two were entirely in Scala/Akka/Spray except for a scriptable CLI that I wrote in Python. I laid out and set up the architecture, wrote most of the infrastructure and significant portions of the management code, and served as "the Scala/Akka/Spray expert" for the rest of the team, who initially knew no Scala.

Yunteq was a tiny startup that made a virtualization management product. I designed and wrote the REST API by which you could talk to it, a scriptable CLI to the API that served as the main user interface, the upgrade system, the API test harness, the console proxy logic, and more. I re-architected the system for the second generation of the product. Eventually Yunteq was acquired by Coraid.

- Helped develop architectural simulators for SPARC systems in C, and prototyped an object-oriented version using C++.- Wrote a tool called Warlock that finds locking problems (deadlocks, data races, etc.) in multi-threaded C code using static analysis. There was a 5000-line C++ component (the analyzer), significant modifications to Sun's ANSI C compiler front end to capture program semantics as input files to the analyzer, and a scriptable CLI (written in Perl).- Worked in various capacities in the Solaris and Java ecosystems.

We put networks of computers into hospital ICUs to keep track of everything happening to patients, from doctors' orders to drug administration to events from monitoring devices. I developed infrastructure software that the other developers used, including a messaging system, a fast library for dealing with a compact date/time representation, and more.

I spent 3 years in the group that created the VersaDOS operating system and hardware (VERSAbus and VMEbus boards, primarily), followed by 3 years in the Government Electronics Division.In the first group I first worked on utilities (e.g. the backup utility) in 68000 assembly language. Later I wrote device drivers, and ultimately ended up leading the kernel group. I prototyped a radically different real-time kernel architecture that was 54 times as fast as the one in VersaDOS at non-blocking I/O, imposed zero overhead on interrupt handling, dispatched processes in constant time, and provided excellent isolation between components. I sometimes worked with board designers to improve performance of embedded systems by suggesting hardware mods.In the Goverment Electronics Division I worked on a variety of different projects, most of them short-lived, including some interesting proposals for government-funded projects. Here I was exposed to the Lucid dataflow programming language and did further work on the real-time kernel prototype, including retargeting it to Motorola's new RISC chip, the 88000. I made several suggestions to the 88000 designers, two of which they incorporated into the design. I also learned some Prolog there, as some team members were working with an alternative to the Warren Abstract Machine. And I ported pSOS to a system the group was building. I was given Secret Clearance.