Portfolio — 2026 · Cambridge, MA

Finn Pounds

Welcome. Here's what I've been working on.

  • M.S. in Analytics, 2026 — concentration in Applied Machine Intelligence
  • B.S. in Mathematics & Physics — Tufts University, 2019
  • MBAA-certified packaging operator — brewery fix-it guy, SOP writer, QC-panel designer
  • Former bike-shop technician
  • Professional kendama player & international freestyle judge

Python · TypeScript · Claude API · PyTorch · YOLOv8 · FastAPI · Next.js

fig. 1 a diffeomorphism of the unit square

fig. 2 — configurations of two and three circles

after Kandinsky · Several Circles, 1926

01 / Selected work

Applied AI, computer vision, and a kendama camera operator.

Things I've built end-to-end: production LLM apps, CV pipelines, and Responsible-AI tooling. A few aren't public yet; links land where they can.

P-01

KendamaVision

End-to-end computer vision for kendama: a fine-tuned YOLOv8 detector + BoT-SORT tracker, wrapped in a physics-aware post-processor — gravity priors, depth-normalized gates, a soft string constraint — that fills in the fast, blurry tama where the detector misses it, then auto-keyframes a "camera-follows-your-trick" clip. On clips it never trained on (as far as my training records show), the detector finds the ken at about half of hand-marked moments and the tama at under half. A SAM2 auto-labeling → retrain loop turns a few taps per clip into training labels. The writeup has interactive demos running on real tracking data.

YOLOv8BoT-SORTSAM2Streamlit
P-02

Kendama Ratings

Unofficial Glicko-2 ratings for competitive kendama, rebuilt from Challonge brackets that event organizers published (NAKO, Battle at the Border, EKC and smaller events so far). Every rating is shown as a range, and players are ranked by its low end, so a high place needs a long record behind it. The Python pipeline ingests brackets, reconciles players entered under handles, accounts and bare first names, and tests itself on held-out events: between players it had already seen, the higher-rated one won about three in four matches. A freestyle adjustment, learned from freestyle battles only, orders the Freestyle board.

Glicko-2Challonge APIPython
P-03

ConceptShape — Semantic Geometry Explorer

An interactive 3D "microscope" for the inside of a language model: watch a sentence's tokens move and cluster layer by layer through the residual stream, compare different models' geometry (CKA + Procrustes), and measure the topology of a corpus with persistent homology — a hands-on take on the Platonic Representation Hypothesis. TransformerLens + PyTorch backend, Next.js + React-Three-Fiber front-end.

TransformerLensFastAPINext.jsR3F
P-04

SkyWatch

A production Claude-API weather monitor. An hourly Python pipeline ingests Open-Meteo, NWS, and HRRR GRIB2 data into derived meteorological variables; Claude turns them into plain-language briefings, with response caching and cost-stratified model selection across Haiku / Sonnet / Opus.

Claude APIFastAPINext.jsGRIB2
P-05

biasface

A demographic-bias audit of off-the-shelf face models: a linear probe recovers race from ArcFace and ViT-Base embeddings at ~0.85–0.90 held-out AUC — far above chance, though neither was trained to encode it. Paired with the compliance stakes under GDPR, CCPA, and the NIST AI RMF. Built for EAI 6400 (AI Ethics & Governance).

PyTorchResponsible AINIST AI RMF
P-06

Encrypted Vitals

Can a model flag a tumor from a patient's vitals without ever seeing them in the clear? A logistic-regression classifier is compiled to a fully-homomorphic-encryption circuit (Concrete ML / Zama) and runs on encrypted vitals — the server computes on ciphertext it can't read, and the encrypted predictions match the plaintext ones bit-for-bit. Built for EAI 6400.

Concrete MLFHEscikit-learn
P-07

GroundTruth

A browser game: classify EuroSAT satellite tiles and race a ResNet18 running live in your browser via ONNX Runtime (WebAssembly) — human vs. AI accuracy, entirely client-side, no server. Built for EAI 6010.

ONNX RuntimeResNet18ReactEuroSAT
P-08

BoneSim

Dominoes against a Monte Carlo opponent. On its turn, the computer plays out 50 random games for every legal move and picks whichever move won most often. There's no hand-written strategy, just simulation, and it all runs in the browser. Built for fun.

Monte CarloNext.jsReact
+

More on GitHub

Public repositories, experiments, and works in progress.

@finnpounds

02 / Academic

Mathematics and physics, now machine intelligence.

B.S. in Math & Physics (Tufts) → M.S. in Analytics with a concentration in Applied Machine Intelligence (Northeastern).

[1]

M.S. in Analytics.

Northeastern University · Applied Machine Intelligence concentration.

Boston, MA · conferred June 2026

[2]

B.S. in Mathematics & Physics.

Tufts University.

Medford, MA · 2019

[3]

CareEscapes: a real-time travel-cost dashboard for medical tourism.

Team project — Finn Pounds et al., ALY6080 Integrated Experiential Learning.

Northeastern University · 2025

[4]

KARA: predicting first-year infant mortality from birth-certificate data.

Finn Pounds, ALY6980 Capstone · industry sponsor Kids at Risk Action (KARA).

Northeastern University · 2026 · XGBoost on 25.8M CDC births (held-out AUROC ≈ 0.93), with a race-stratified fairness audit and state-level disparities

Selected graduate coursework

  • EAI 6400 Data Governance & Responsible AI
  • EAI 6010 Applications of Artificial Intelligence
  • EAI 6020 AI System Technologies
  • ALY 6030 Data Warehousing & SQL
  • ALY 6040 Data Mining Applications
  • ALY 6110 Data Management & Big Data
  • ALY 6140 Analytics Systems Technology

03 / Kendama

Judging, sponsorship, and a few clips.

Freestyle judge and sponsored player, mostly around Boston and the U.S. contest circuit.

Judging

FISE — International Freestyle Judge

The American judge, with Tamotsu (Gloken) & Teva (One Kendama)

European Kendama Championships

Judge — prelims (Leiden, Jan) & finals (Utrecht, May) · 2026

Battle at the Border 2026

Freestyle judge · Nashville, TN

Recap ↗

NAKO — North American Kendama Open

Judge · 2025 & 2026 · Minneapolis

…and Fair Games, the United States Kendama Open, and the Nordic Kendama Cup.

Sponsorship

Kendama Depot

Sponsored player · current

On the team — kendamadepot.com · welcome edit ↗

Paradox (PDOX)

Professional kendama player & co-founder · current

A kendama brand I co-founded; designed a signature ken shape that's sold out multiple times. 18pdox18.com ↗

KROM Kendama

Brand Ambassador · previous

kromkendama.com ↗

Community

mass.kendama

Local jams · Boston

Helping grow the Massachusetts kendama scene. @mass.kendama ↗

@finnpounds

Instagram

Follow along ↗

04 / Bikes

Built, ridden, repaired.

Avid builder and shop-trained repair tech — every one of these is a custom build, frame up.

05 / About

A short note, and one on the artwork.

Hi, I'm Finn. I live in Cambridge, studied math and physics at Tufts, and recently finished an M.S. in Analytics (Applied Machine Intelligence) at Northeastern. Before that I fixed bikes in a shop and kept a brewery's packaging line running, and I still like work where something real is broken and needs fixing. Lately that's messy data: I like turning it into tools that help people make a decision, often with a language model somewhere in the loop.

Away from the keyboard I play and judge competitive kendama, including the freestyle panels at Battle at the Border and the European Kendama Championships, and am sponsored by Kendama Depot and Paradox.

If you're hiring around applied ML, data tooling, or LLM products, or you just want to talk kendama, say hello.