I grew up in Milwaukee, WI with music and video games. After getting my BS in Mechanical Engineering at MIT in 2017, I moved to Palo Alto, CA. I worked for Kitty Hawk as a controls engineer. I worked on various control systems and software for a Vertical Takeoff and Landing (VTOL) electric aircraft that cemented my love of flying and controls. It also introduced me to Haskell and the greater functional programming space, which I am reluctant to leave. I am now pursuing my Ph.D. at the University of Pennsylvania under Dr. Vijay Kumar and Dr. Mark Yim.
As an amateur, I am currently learning guitar, photography, and eventually mountaineering. Past projects I wish I had more time for include photography, BattleBots(tm), electric vehicles, power electronics, classical guitar, violin, piano, kites, hiking, rock climbing, web design, jiu jitsu, and painting.
Random Project Ideas
This is an ongoing list of completely random project ideas unrelated to my professional work. If you happen to build one of these, let me know! I would love to see it.
- The Modern Illustrated Primer. In Neal Stephenson’s The Diamond Age, a young girl encounters a “magic book” which engages her, responds to her questions about matter and life, and ultimately trains her in everything from martial arts to nanotechnology to being “clever”. This shows a deep parallel with Massively Open Online Courses (MOOCs) in that vast amounts of information is stored at your fingertips, ready to be learned by you. However MOOCs have extremely poor levels of completion. According to HarvardX, only 22% of people who go in wanting to get a certificate actually do so. What if a MOOC could talk back to you, have a conversation, explore what you enjoyed and relate it to your life? Perhaps with modern machine learning this is not so far away…
- Augmented Depth Perception using Machine Learning. RGB image to Depth Map machine learning algorithms are pretty common these days. Some of them give back uncertainty about the depth map pixels. Using that uncertainty, you could augment the neural network part with
- Handgun DSLR. Modern DSLRs are somewhat silly shaped. The true shape of a modern DSLR should be a handgun, much like Fluke’s thermal cameras. It is much more ideal as it is a more ergonomic and stable way to hold a camera, as well as something that might be easily holstered compared to other cameras. Foreseeable problems could be that with long lenses, the weight balancing would be incorrect. However the current weight balancing is already incorrect.
- Rustkell. A generic robotics platform such as ROS using modern languages. In particular, the C/C++ low-level and embedded is replaced with Rust, while the higher level logic is Haskell. Strong typing, memory safety, and aggressive compilers among other things would hopefully ensure that Ruskell would be a framework oriented around runtime safety, efficiency and clarity.
- Coffee Grinder Alarm clock. Coffee smells good to basically everyone. For addicts, it has the added benefit of psychologically triggering the feeling of being awake. A coffee grinder that propels the scent of coffee into the air as one wakes up, combined with the loud and annoying sound of grinding would be a smash hit.
- Electric skateboard “cruise control”. Two deadman switches are on the board. The front switch, when depressed, releases the electric brake, such that if you fall off the board, the board brakes and doesn’t run away. The back switch turns the board from “maintain speed” to “coast”. The idea is that the rider kicks up to a certain speed (front foot engaged so brake is off). When the desired speed is achieved, the rider puts their foot on the back switch, telling the board to maintain that speed. If they want to slow down slightly, the rider moves their back foot off the back switch to change to coast, and if they need to slow down quickly they remove their front foot coming to a stop (likely more abruptly than ideally desired).
- Flamingo leg model. According to this paper, flamingos are stable while resting on one leg but not on two. Dead flamingos were used to determine this. I would like to create a physical anatomical model to understand the legs.
- Machine learning takes forever. Build an inverted pendulum (bonus points: double pendulum) and set a machine learning algorithm out on it to learn how to balance it. Alternatively can switch to a pre-written controls code. Watch as the machine frustrates with the swing up and basic control. Good desk toy.