Dylan Glatt

Marketing Mix Model Version Control

Ipsos MMA

Marketing mix models shared across international offices were constantly crashing — no version control, hardcoded formulas, inconsistent documentation. I created structured Excel templates that centralized assumptions and moved collaboration to dedicated Microsoft Teams channels. Analytics model crashes were eliminated.

GM: 48-Model Delivery

Ipsos MMA

General Motors has 48 nameplates — each requiring its own marketing mix model. The engagement was the largest our team had handled. As part of the US delivery team, I organized the data collection, validation, and QA processes that made delivery effective across all 48 marketing mix models.

LLM-Powered Song Analyzer

riffd

A single tool that separates the instruments, identifies the key, and tells you what the chord progression actually is. As a musician, I found nothing that could take a real song and explain the theory behind it — so I built one from scratch, directing Claude as the implementation layer, and deployed it with Flask.

fitness-agent

WHOOP knows recovery, Strava knows workouts, training plans live in your head — none of them tell you what to do today. I built a personal AI coach that lives in Discord, fuses overnight physiology with activity history, weather, and a training-literature RAG, and prescribes a specific session every morning the moment recovery lands.

Real-Time Event Intelligence

Argus

The data intelligence agencies use is publicly available — events, filtered and verified, stripped of noise. It’s just buried under hundreds of thousands of unverified daily reports. I built the filtering and verification pipeline from scratch using public sources and the Palantir open source framework.