ian@pentland:~$
[   0.000000] booting portfolio v2.0
[   0.041233] mounting /home/ian ... OK
[   0.097114] ready.

IAN
PENTLAND_

Sophomore @ WashU — joint Business + Computer Science.

./projects github ↗
>_
// 01 — about
ian@washu
PROGRAM... Business + CS (BUCS)
LOCATION.. St. Louis, MO
CLASS..... 2029
ROLE...... Product Space Fellow

I am passionate about data analytics and technology. I believe that their intersection advances society through solving global issues, optimizing individuals' lifestyles, and providing meaningful insights.

// 02 — projects
polyedge [active] MONEY · DATA PLATFORM

POLYEDGE

A data platform for prediction markets. It pulls in 22M+ trades across 400K+ markets and stores them so I can test trading ideas against real history instead of guessing.

My first idea for an edge didn't hold up when I tested it, so I moved on. The next one did — a pricing gap worth about 5.8¢ per $1 contract, which held up across 21K simulated trades. One of the 900+ automated tests caught a data bug before it could skew results.

Python DuckDB Parquet Linux
view on github ↗
22M+trades stored and analyzed
+5.8¢edge per $1 contract, after costs
7 / 8quarters it stayed profitable
900+automated tests behind it
llm-from-scratch [complete] AI · MACHINE LEARNING

LLM FROM SCRATCH

A small language model, built and trained from scratch in PyTorch on 1.35B tokens — no pre-built model, just the underlying pieces (attention, tokenizing, training loop) put together by hand.

I then pulled 96K+ research papers and fine-tuned the model on them, improving its accuracy on that material by 32.6% — at the cost of being a bit worse at everyday language, which felt worth being honest about.

Python PyTorch Transformers
view on github ↗
162Mparameters, built from scratch
96K+papers used to fine-tune it
32.6%more accurate on domain text
ai_ear [in progress] AI · AUDIO SIGNAL PROCESSING

AI EAR

Allowing AI to hear music. AI's frequent text patterns, such as adding suspense or drastic juxtapositions, makes it seem like AI would interpret music with some form of false "emotion."

Rather than describing the music, this script transforms mp3s and m4as into frames of representation of the sound waves.

Python librosa NumPy FFmpeg
view on github ↗
144frequency bins per frame
86 fpsacoustic frames per second
9octaves, 33 Hz to 16.8 kHz
0semantic labels in the output
SIDE HUSTLE

Custom PC Building

~$10K in builds for classmates, ~$2K profit — plus the Linux server that runs Polyedge.

PRODUCT

Product Space × Microsoft

A fellowship where I talked to users and researched how to make AI tools easier to use.

CASE COMPETITION

Lyft Case Competition

Placed in the top 5% presenting to judges from Olin Business School.

// 03 — experience
Expected
May 2029

B.S. Business + Computer Science

Washington University in St. Louis

BUCS joint program (2% acceptance rate) · 3.7 GPA · Dean's List ×2 · $30K merit scholarship · 2nd place, WashU Startup Pitch Competition ($1,250).

Oct 2024 –
Jan 2025

AI Research Assistant

University of Alabama at Birmingham

Analyzed CDC public-health data with Dr. Yusen Zhai to study how social factors influence mental-health outcomes.

Aug 2024 –
May 2025

AP CS A Teaching Assistant

RSU21 · Kennebunk, ME

Taught Java fundamentals, debugging, and OOP to 20+ students as an interim educator; led labs and reviewed assignments.

// 04 — skills
use all the time working knowledge

LANGUAGES/

python java sql javascript html/css

TOOLS/

pytorch duckdb pytest · git linux/systemd parquet · mypy · ruff

AREAS/

data pipelines backtesting model eval · REST APIs transformers automation
// 05 — contact

SAY HI.

Open to internships, research, and interesting side projects.

i.pentland@wustl.edu linkedin ↗ github ↗
© 2026 ian pentland