Mathematics · Computer Science · Union College ’28

Syed Mahir
Ahamed

I’m a math and computer science student at Union College, mostly interested in where AI and mathematics run into each other. I’ve tried my hand at a few problems in that space, and I’m still working out which corner suits me.

See my research Get in touch

01 / About

A little about me.

I’m studying math and computer science at Union College, with a minor in statistics, and I’ll graduate in 2028. Most of what I work on sits between the two subjects. I like questions where a neat mathematical idea has to actually hold up once it meets real data or a real system.

I’ve been lucky to try that out in a few pretty different settings so far: reinforcement learning theory, a study on how people read AI-written text, some numerical linear algebra, and a bit of pure algebra. None of it is finished, and I’m not an expert in any one of them yet, but each has been a good excuse to learn something new. On the side, I help other students with research methods as a peer mentor at our library.

Some coursework

  • Abstract Algebra
  • Linear Algebra
  • Numerical Analysis
  • Mathematical Statistics & Probability
  • Data Structures
  • Object-Oriented Programming
  • Large-Scale Software Design

Tools I use

  • Python
  • Java
  • NumPy
  • SciPy
  • statsmodels
  • PyTorch
  • Git
  • LaTeX
  • SPSS
  • Macaulay2

02 / Research

Things I’m working on.

A handful of projects across AI, statistics, and math. They’re at different stages, and most are still going. Happy to talk about any of them.

Accepted · RLC 2026

Infra‑Bayesian Reinforcement Learning

SPAR AI (Supervised Program for Alignment Research)

I’ve been looking at how reinforcement learning agents make decisions when the world is adversarial, or when they can’t fully pin down what’s true. We set up small Newcomb-style problems to test the ideas, and put together a shared codebase with a toy infra-Bayesian agent. It’s early work, but a short paper came out of it that was accepted to a workshop at RLC 2026.

  • RL theory
  • infra-Bayesianism
  • AI safety
Read the preprint ↗
In progress

How People Spot AI‑Written Text

Union College · CAPLAB

A study of how well people can tell whether something was written by a person or a model, and what gives it away. I put together the analysis pipeline (a mixed-effects logistic regression in Python and R) and ran it across four datasets, and I helped run the follow-up interviews where people talked through their guesses. One thing that stuck with me: AI creative writing got flagged around 84% of the time, while judging plain factual accuracy was closer to a coin flip. We’re writing it up for CHI.

  • HCI
  • mixed methods
  • AI text detection
In progress

Randomized SVD & Low‑Rank Approximation

Union College · Scholar’s Project

My Scholar’s Project, on using a bit of randomness to approximate large matrices cheaply. I implemented randomized SVD and compared its accuracy and speed against the usual methods like power and QR iteration, with the Eckart-Young theorem as the reference point. I checked the low-rank reconstructions on real images, where the error lined up with what the singular values predicted. I’m slowly working it toward a manuscript for a SIAM venue.

  • numerical linear algebra
  • SVD
  • low-rank approximation
In progress

Commuting Matrix Varieties

Union College · Mathematics

A more pure-math project, poking at the combinatorial structure of commuting matrix varieties, the pairs of matrices A and B where AB = BA. I’m using Macaulay2 for the computer-algebra side and writing small packages to take care of some of the routine computations.

  • algebra
  • algebraic geometry
  • Macaulay2
2nd place · AIxBio

BioRT‑Bench

Apart Research · AIxBio Hackathon

A weekend project with a small team: a red-teaming benchmark for checking how well frontier language models hold up against misuse prompts in biosecurity. We tried four attack methods against four models across a set of prompts, and scored the responses with a calibrated judge adapted from StrongREJECT. It ended up placing second at the hackathon.

  • AI safety
  • red-teaming
  • evaluations

03 / Resume

Resume

If you’d rather see everything on one page, here’s the short version: education, research, and the tools I use.

View resume (PDF)

04 / Contact

Say hello.

If you want to talk about any of this, have a research opportunity, or just have a problem you think I’d find fun, email is the easiest way to reach me.