Neuroscience @ Pitt · Frederick Honors Neurotech research @ CMU

I study brains and build software to explain them.

I'm Ben, a neuroscience student who builds AI tools. My research models movement and the nervous system; my side projects make AI more transparent, from mapping every assumption a coding model makes to a notebook that checks your math step by step.

See the work ↓ GitHub ↗ LinkedIn ↗ Resume ↓

hover to trace · click to jump

researchbuildsskills

Selected work

software · AI · tools I use daily

MindCodeNov 2026NOVA hackathon · ScottyLabs, CMUTeam lead

Every assumption an AI makes, as a map you can edit before it builds.

When AI builds an app, it silently decides dozens of things you never asked for. MindCode turns your idea into a full plan: pages, features, APIs, design, audience and cost, colored by what you said vs. what Claude assumed vs. what's still open. Then it builds the app and traces every element back to the bubble that caused it.

Designed the prompt-engineering framework (Design / Audience / Functionality / Cost) that structures every plan.
Led 4 people and a team of AI coding agents (Claude Code, Google Antigravity, Jules) through a shared task board, cross-review and a 15-deploy budget.
Right-click any part of the generated site to change it (~7 s, via patch edits); small plan changes rebuild in ~20 s.

See the app it built ↗Try MindCode ↗Code ↗

Ben's Notes2026TypeScript · PWA · 72 commitsSolo

A notebook that reviews your notes as you write them.

Ben's Notes (also called Pipit Notes) is an iPad notebook wired to machine-learning models. While you write with Apple Pencil, it reads your handwriting, checks it, and turns it into study material built on learning-science methods that have been tested in research. Notes stay on your device in IndexedDB, with no account and no server.

Live review as you write: handwriting is recognized into math and text, each line of working is checked (SymPy running in the browser through Pyodide verifies the algebra), and mistakes are highlighted on the page with why they're wrong and the fix.
Exams that look like your real ones: timed practice papers generated from your notes and course PDFs, with structured multi-part questions, graphs, tables and diagrams cut from your own materials, marked against a marking scheme.
Proven study methods: FSRS-5 spaced repetition (the scheduler Anki uses), retrieval practice, interleaved short lessons, and concept maps you recall from memory and that are colored by how well you know each idea.

Open the app ↗Code ↗

Pipit Finance2026Next.js · Prisma · PostgresSolo · private

A full personal-finance platform, connected to real bank accounts.

A budgeting and net-worth app in the spirit of Monarch or Copilot, built end to end: about 440 files, 38 database models and 88 commits. Bank accounts sync through Plaid and subscriptions bill through Stripe.

Bank sync: Plaid Link and incremental transaction sync, with signature-verified webhooks (JWT). Access tokens are encrypted at rest with AES-256-GCM.
Security: NextAuth sessions, TOTP two-factor auth with QR enrolment, rate limiting, and Zod validation on server input.
Money logic: a rules engine and auto-categorizer, recurring-charge detection, budgets, shared household goals, investment history, cash-balance forecasts and net-worth projections.
Reports: Sankey cash-flow diagrams (d3-sankey), spending reports, salary benchmarks, CSV import and export, and scheduled snapshot and alert jobs with email through Resend.

Private repo: code walkthrough on request

The Ben Times front page: lead story, weather, homework, the week ahead and flagged emails (personal details blurred)

The Ben Times puzzles page with a word-ladder game open over a mini crossword and comic strip

The Ben TimesPythonprivate

A personal morning newspaper, with puzzles.

A scheduled job reads my Gmail, Calendar, Notion and the weather; Claude writes the copy; it renders as a two-page broadsheet for an old iPad on my desk. It ends with a puzzles page (mini crossword, word ladder, jumble) that's playable on the iPad, and it falls back to a plain edition when the writer fails its checks.

NFL EdgePythonprivate

Predicting games, tested against Vegas.

A model of NFL game outcomes, backtested against historical betting lines to see where it would have found an edge, and where it wouldn't.

Agent boardBash · GitHubopen source

Getting AI agents to work as a team.

A shared GitHub-issue "board" that lets several coding agents claim tasks, tag each other, review each other's PRs and hand off work. It's the system I used to run MindCode's build.

See the real board ↗

Research

neuroscience · ML · diagnostics

Parkinson's · MLUT Austin · 2025

Predicting freezing of gait before it happens.

An ML framework that predicts Freezing of Gait in Parkinson's seconds before onset, from wearable accelerometer data (Daphnet). 74 engineered biomechanical features; XGBoost, CatBoost and deep sequence models under leave-one-subject-out validation; SHAP to check the predictors are physiologically meaningful.

90%recall

82%F1 (XGBoost)

Regeneron Biomedical Science Award and 1st place, Greater Austin Regional Science Fair. Acknowledged contributor, British Journal of Multidisciplinary and Advanced Studies (2025).

NeurotechnologyCMU · Beauchamp Lab · 2026–now

Listening to motor neurons through the skin.

Recruited as a Pitt freshman to Dr. Drew Beauchamp's lab (Biomedical Engineering and the Neuroscience Institute). I decompose high-density surface EMG in MATLAB into individual motor-unit firing patterns, to understand how spinal motoneurons relate to signals from the cortex and subcortex.

The lab ↗

DiagnosticsBabson Diagnostics · 2025–26

First high-school intern in company history.

R&D scientist intern across wet and dry lab work: built reference ranges from real patient data, designed experiments for neuro and testosterone assays, wrote SOPs, and supported the data-to-FDA pipeline.

700kpatient samples analyzed

−20%assay waste

Experience

research · industry · leadership

Aug 2026 — now

Neurotechnology Researcher · Carnegie Mellon, Beauchamp Lab

High-density surface EMG decomposition and motor-unit analysis in MATLAB.

2026 — now

Founder · Kadison Research

Pre-med and biotech mentorship for high school students navigating research, internships and pre-med paths.

May 2025 — Jul 2026

R&D Scientist Intern · Babson Diagnostics, Austin

Reference ranges from 700k+ samples, −20% assay waste, deliverables for two Fortune 500 partners, FDA data pipeline. Rural-health research for the board contributed to a $50K state initiative award.

May — Jul 2025

Research Fellow · UT Austin High School Research Academy

Freezing-of-Gait prediction in Parkinson's with wearable sensors and ML (mentor: Prof. Patrick Benfield).

Jul 2025

Emergency Relief Coordinator · Central Texas floods

Coordinated logistics, volunteer teams and thousands of donated supplies for displaced families.

Dec 2024

RAC Advocacy Fellow · Washington, D.C.

Co-authored and delivered a legislative advocacy speech to members of Congress.

Lately on GitHub

live from the GitHub API

Nov 2026Shipped MindCode's final release at NOVAMindCode Nov 2026Ben's Notes: graphing, practice exams, FSRSpipit-notes

About

I'm studying neuroscience (bioengineering minor) at the University of Pittsburgh in the Frederick Honors College. I'm drawn to the place where biology meets computation: decoding what the nervous system is doing, and building tools that make complex systems, from muscles to AI models, easier to understand.

Outside the lab I build things I actually use every day, I mentor students through research and pre-med paths, and I'll organize a relief effort when my community needs one.

Computational

PythonTypeScriptMATLABPandas · scikit-learnXGBoost · SHAPTime-series & EMGLLM apps · AI agentsReact · Netlify

Laboratory

Assay developmentExperimental designSOP authoringClinical dataGel electrophoresisMicropipetting

Certifications

Python for Data Science, AI & Dev · IBMMedical Terminology I–III · RiceHealth IT Fundamentals · Johns Hopkins