A study in finding what works. Boston, MA

The optimal point found fast.

I’m William Balz. My work moves between pure math, advanced manufacturing, and early-stage software. Different fields. The same instinct: find the answer hidden in the possibilities.

Fig. 001The possibility space

MANY POSSIBILITIES. ONE DIRECTION.
DRAG TO ROTATE · REPLAY TO CONVERGE

Electrical & Biomedical EngineeringTufts University · Class of 2029

SCROLL TO EXPLORE

An old theorem.
A new way to see it.

Mathematics / 2025
Published in Parabola

FIG. 002 / CONVEX KINEMATICS
Ω Γ p q [Ω] − [Γ] = πpq
p : q = 50 : 50

A circular case of a surprisingly universal result.

Published research

On Convex Kinematics:
A Resurrection of Holditch’s Theorem

Holditch's Theorem is a strange, old fact about convex shapes: slide a fixed-length chord all the way around the boundary of any convex region, and a point riding along that chord traces out a smaller region. However oddly the outer shape is drawn, the two areas always relate by one clean formula. Reverend Holditch's original 1857 proof runs about ten lines but explains almost nothing about why it's true. Sawyer and I rebuilt the result on more concrete ground instead — using Green's Theorem to handle area along a curve, and an unexpected triangle, tied together by Stewart's Theorem, to make the whole construction visible step by step.

Explore the proof (PDF)

Written over Thanksgiving break in 2024. Reviewed by Dr. Bill Dunham at Bryn Mawr College before publication submission.

Less theory.
More making things.

A few things
I’ve built.

After Hours: an illustrated museum at night, with specimen galleries, security beams, and a locked Darwin vault.
HANDSHAKE GAME CHALLENGE 01 / MULTIPLAYER

After Hours

A four-player museum heist where knowing a little science can get you a long way. Race through three museums, experiment with stolen specimens, and figure out what belongs together before someone opens the vault. I built it with ChatGPT for Handshake’s game challenge, exploring how quick movement and careful deduction could share the same game.

React / Canvas Realtime multiplayer Desktop + mobile
PLASTIBOT / SPECTRAL SIGNATURE
85% ACCURACY
06 MATERIALS / 09 COLORS
TUFTS IDEA LAB 02 / MACHINE LEARNING

Plastibot

A hyperspectral-imaging project that sorts plastic waste by type, aimed at higher-quality recycling. I lead the team and built the current classifier — a convolutional neural network in Keras that identifies six plastic types across nine colors at 85% accuracy, about 15 points better than last year's model.

Hyperspectral imaging Keras / CNN Team lead
SUSHI GO / A LITTLE FRIENDLY COMPETITION
02
●
MAKI ROLL
03
✳
SALMON NIGIRI
01
◆
DUMPLING
C++ / HEURISTIC SEARCH
SIDE PROJECT 03 / GAME AI

Sushi Go bot

A small C++ bot for the card-drafting game Sushi Go. Less serious than it sounds — mostly an excuse to see how far a simple heuristic search gets inside a game small enough to reason about by hand.

C++ Heuristic search Just for fun

There’s always another experiment.

More on GitHub

A different
kind of growth.

Since January 2026 I've been an undergraduate researcher in Tufts' cellular agriculture lab, working on genetic modifications and media formulations aimed at faster cell growth, better suspension viability, and greater cell density. I also train new lab members on protocol and sterile technique. Right now, I'm most excited about our new de novo lipogenesis project, which we plan to attack from several different biological angles.

Cellular agriculture · The Stout Lab
William Balz
WILLIAM BALZ BOSTON, MA

Work first.
Credentials later.

I'm a sophomore studying Electrical Engineering and Biomedical Engineering at Tufts University, after a high school career built around moving fast: I taught myself a full year of Calculus BC in under a month and walked into the AP exam cold, scoring a 5, later taking Linear Algebra and Multivariable Calculus at Harvard. I also took junior-level EE Probability & Statistics at Tufts as the only freshman in the room.

That same instinct — work first, credentials later — carried into everywhere I've worked since: optimizing manufacturing processes in metal 3D-printing at VulcanForms, founding an ed-tech startup from inside a venture fund at Morningside, and mentoring new researchers at Tufts' cellular agriculture lab. It started, though, with a proof.

“I remember sitting at my desk on the phone with my friend during our Thanksgiving break. It was the 40th hour that week we'd spent trying to prove Holditch's Theorem in a more intuitive, concretely geometric format. We were sold on using Green's Theorem, but hadn't made much progress until a last-ditch effort late on Sunday night: an implicit triangle that would tie everything together with Stewart's Theorem.”

— My sappy recounting of the creative process

The same pattern showed up again at VulcanForms, when a mentor left for three days and asked for an optimized laser-marking protocol by the time he got back — thousands of settings, no obvious way to test them all. And again at Morningside, when a stalled prototype turned into two days of learning Python and building the thing myself. Different fields, same move: narrow an enormous space of options down to the one that actually works, fast.

GOOD QUESTIONS WELCOME.

Let’s find
what’s next.

Happy to talk about math, projects, or anything adjacent to search, optimization, or making things work under a deadline.