Chris Levy

Dash Social

Formerly Dash Hudson · July 2019–present

1. Staff AI Scientist · July 2025–present
2. Senior Research Scientist · July 2022–July 2025
3. Research Scientist · July 2019–July 2022

Generative media & creative agents

Built generative AI workflows for on-brand social media assets using image and video generation APIs. Created a Brand DNA pipeline that segments brand feeds and adds searchable image descriptions, enabling agents to retrieve relevant visual inspiration. Developed a conversational creative agent to orchestrate generation and editing, with custom evaluations for product fidelity, text accuracy, and brand consistency. Partnered closely with product teams, using working prototypes to demonstrate capabilities, align on product direction, and give engineering a foundation to build on.

Conversational analytics & coding agents

Built analytics agents using function calling to retrieve product data. Evolved these into stateful coding agents using Modal sandboxes for arbitrary code execution and data analysis. Developed evaluations for tool use, numerical accuracy, and response quality, and worked with product and engineering teams to prototype and ship the systems.

Encoder fine-tuning & serving at scale

Fine-tuned transformer encoders for classification tasks including creator interests, location prediction, multilingual sentiment, and language detection. Distilled larger decoder LLMs into compact encoders for faster inference. Built labeling, training, evaluation, and production serving pipelines to run these classifiers at scale.

Visual trend discovery for Social Listening

Built the visual-trend discovery pipeline for Social Listening, grouping social posts into visual themes by clustering ResNet-derived image features. Integrated Elasticsearch retrieval, duplicate and outlier filtering, and cluster-level engagement and sentiment summaries, and deployed the pipeline as a background service.

Computer vision

Developed computer-vision systems for large-scale image deduplication, visual retrieval, and engagement prediction. Built celebrity recognition using FAISS k-nearest-neighbor search over face embeddings. Integrated image embeddings into Elasticsearch’s inverted index using custom encodings for visual similarity search and duplicate-candidate retrieval. Trained classifiers for face and text presence, fine-tuned product detectors, and developed custom product-search embeddings. Led major refactors from legacy TensorFlow 1 to TensorFlow 2, then migrated the model suite to PyTorch and optimized inference.

Earlier experience

Data Scientist · Mobivity
January 2017–July 2019

Data Scientist · Morneau Shepell
January–December 2016

I've created several Udemy courses on SQL, differential equations, Python, and AI engineering, reaching 32,000+ learners and receiving 3,800+ reviews.

Previously, I was an Instructor at Dalhousie University (2013–2016), primarily teaching Differential Equations I and II. I also taught calculus and mathematics for commerce, and helped develop online assignments and course material.

Explore my courses on Udemy ↗

PhD · Applied Mathematics

Dalhousie University · 2015

Thesis: Modelling the Spatial Effects of the Signal Transduction Process

Scholarships and awards: NSERC Alexander Graham Bell CGS-D, Killam Predoctoral Scholarship, and Dalhousie President’s Award.

MSc · Applied Mathematics

Dalhousie University · 2010

NSERC Canada Graduate Scholarship (CGS-M).

BSc · Mathematics

Mount Saint Vincent University · 2008

Awarded Highest Aggregate in the Bachelor of Science (Honours) Program.