Shaurya Singh · Seattle, WA
cs + data science at UW, class of '27.

everyone sensible says pick a lane, and i keep not doing it. so far that has meant proteins, brains, quantum noise, classrooms, street safety, buildings, an underwater robot, and the code that grades things.

$ grep -rl "same field" ~/built/
no matches

one line per world i've wandered into.

• proteins · undergrad research in UW's DAIS lab: preprocessing, training runs and renders for DeepTracer, which turns cryo-EM maps into 3D protein structures. no byline, and that's fine.
• brains · taught myself neuroscience in public: four essays, optogenetics to connectomics (2022–23). lately, trigalign, for EEG recordings that dropped a trigger or two.
• quantum · shotfloor: the score a perfect quantum computer gets from shot noise alone. 8 qubits, uniform target, 1,024 shots: 0.92 hellinger fidelity, not 1.
• classrooms · co-founded SkillTern and was its CTO (2022–23): computer science for 350+ students, built to run without me.
• street safety · TurtleShell: K-means over 10,000+ LAPD crime datapoints and an iOS SOS app in Swift. got into Microsoft for Startups, then i shut it down on purpose (oct 2023).
• buildings · a cabin-design prototype that has to pass the Seattle residential code: rule engine, NSGA-II over cost, space and comfort, IFC export. runs locally, so no link.
• earlier · H2OAquatics, an autonomous underwater vehicle aimed at ocean acidification, the first time the build-at-it reflex showed up. also a co-written policy paper on phishing, runner-up at WatGov (Waterloo, 2023).
• graders · scoring code that hands out credit it shouldn't. it turned into a side quest; notes at the bottom.
• off the clock · music. made often, released never. links when that changes, not before.

small python tools, each about one thing that quietly goes wrong. none is public yet, so no links: each name becomes one the day its repository is.

tests and repos

• tolwatch · a pytest plugin that flags allclose, isclose and approx checks that can't fail, or are far looser than they need to be.
• ziptrace · names the zip() and map() calls that silently dropped data because their inputs had different lengths.
• unadded · finds the untracked or gitignored files a passing test run quietly depends on, before CI does.
• whymodified · git status says a file is modified and nothing in it changed. this says why, and the one command that clears it.

science data

• shotfloor · the score a flawless quantum computer gets at a given shot count, and how many shots it needs to reach a target.
• trigalign · pairs a stimulus log with an EEG trigger list when triggers were dropped or duplicated, or the clocks drift.

grading code

• scoreprobe · feeds a grader the inputs known to slip through (a blank answer, "not Paris" against "Paris", 4200 against 42) and reports which earned credit.
• normladder · is "1,000" the same answer as "1000"? 8 of 14 answer-matching rules from lm-eval, HELM, SQuAD and friends say yes. it shows which, and why.
• mcqlint · lints multiple-choice eval sets: keys that name no option, duplicate options, leaked answers, position bias.
• tabverify · recomputes the averages, deltas and bold in results tables at the precision they were printed, and flags what no rounding explains.

the habit: i read the code that decides whether an answer is right (eval harnesses, classroom autograders, forecasting and fairness metrics) and file what i find. since 2026-09-23: 238 issues in 229 repositories, 20 fixed by a merged pull request so far, 17 of those by other people and 3 by me.

one example, small enough to read: scoreprobe on a two-line grader, before and after a one-line fix.

_{side-quest numbers: github api, as of 2026-10-08 20:50 utc.}