Some projects

Large-scale Endeavors

/ Orbit

cofounder, 18yo
built neurotech in an abandoned shipping yard.
  • built out new method of neuro-stim (ie. 'writing' to the brain, instead of 'reading' from it)
  • showed some VCs what it could do. raised $10mil seed at $45mil off our tech breakthroughs
  • built a team around the mission of saving the world. built technology I'm proud of.

/ Gamut

founder, 17yo
rallied the gang, and had a blast building cool ML tech with my friends.
  • founded at 17yo. acquired a few months later.
  • 12 patents created on our tech during the acquisition process
  • our answer to: "what should we do with the summer before college?"

/ Deep Protein Models

17yo, @ Mantra Bio
went full research-mode, and got state-of-the-art in the protein-function prediction problem ("given a protein, what does it actually do in the body?"). came up with a wacky new way to featurize proteins (w. graph-neural nets and persistent homology) that powered our ML models. (I learned way later that the medicine-candidate designed w. our system beat out 100s designed by actual biologists to get chosen for in vivo trials :)

/ Condution

cofounder, 14yo
open-source task management app with 4 million+ downloads.
  • 9th-grade quarantine project w. the gang
  • spent my days programming on discord screenshares instead of paying attention to zoom english class
  • we chose the most boring project idea ever, but had soo much fun. led to the default friend-group activity being "build stuff together" :)

/ Math funzies

nerd
chosen to lead the US team in the COMAP International Mathematical Modeling Challenge. #1 USA
  • i'm normally not a fan of competitions, but decided to tryhard this one in an attempt to impress a girl.
  • it didn't go very well.. but at least we won nationals! 🙃
  • also a huge math nerd — grew up on linear and abstract, now a big fan of most things algebra-flavored
    (highlights so far include galois theory, cohomology, spectral graph theory, a tad bit of random matrix theory, info theory. suck at NT and real analysis though..)

Some other random stuff

small huxley builds: got handed a drill press and a chopsaw at 7yo. from 8-10yo, built 40amp levitating electromagnets using electrodynamic suspension, super-corroding alloys to rapidly produce fuel for hydrogen-powered cars, windmills that were dirt cheap to produce by using tensegrity structures, and a bunch of random other stuff.
  • somehow lost zero fingers. (lowkey no idea why my parents were ok just leaving me alone w. power tools 🤷)
electrodynamic_suspension
super_corroding_alloy
tensegrity_stuffs
COVID surge early-warning models @ Mass General: so we could move resources in advance to hospitals that were about to get hit the hardest.
SOCO: seeing with our ears, via iPhone lidar → convert into soundscape → airpod spatial audio. demo vid of albert high-fiving people and navigating while blindfolded.
discovering new chaotic attractors: turns out, it's not that hard! here's one
👇
ANTIDOTE: Artificial Neural network TrojanIning DetectiOn using TDA Estimators (we tried realll hard on the acronym, dont @ me). SOTA on detecting poisoned NNs by looking at the topological properties of a graph grown out of neural correlations.
vietoris_rips
ncg
hebbian
(also tryharded these figures :)
taproot: a shared zettlekesten and note knowledge-base between a bunch of friends. connect to each other's thoughts like links on wikipedia. vulcan mind-meld our second-brains! 30K+ shared nodes, got to present at JMM
no: like GNU yes, but blazingly fast 🚀. (had to learn some assembly for this one)
langdist: solves the language equivalency problem by creating a similarity metric on languages. done primarily through average tree-edit-distance on syntactic structure
2 days
sneaky models: built a model that can encode hidden messages in its output, and then decode and reason over these hidden messages from an entirely separate instance
1.5ish? days,w.@aidan
art: sometimes i do art things. did a lot of video editing in hs
kotae: our pass at a digital whiteboard. infinite canvas, latex w. desmos-style inputs, lets you define / evaluate / operate on symbols, automagically simplifies / expands / solves equations.
epideODE: simulating novel mammalian embryogenesis models
speculative diffusion: like speculative decoding, but switch the fast model for a diffusion model for more speedups, which the heavy model verifies in parallel. demo vid, writeup, code.
and a bunch of random other stuff. hopefully to be added soon :)