Justin Barry
ML research engineer & applied scientist building intelligent systems
Former Applied Scientist @ Amazon • MS Computer Science • BS Mathematics & Computer Science • BJJ Black Belt
I design and implement machine learning systems—generative and discriminative—owning both the math and the production-ready PyTorch code. I work across modalities (vision, language, structured data) to build and adapt architectures from first principles, not just fine-tune whatever API is trending, and get them ready for production.
I'm most valuable when off-the-shelf models or basic baselines stop working. I collapse messy, high-ambiguity problems into clear objectives, then turn a diagram and a loss function into a system you can ship.
I can help your team:
- Own an ML product end-to-end – from problem definition and data, through modeling and experimentation, to metrics, deployment, and iteration.
- Build and extend advanced models – GPT-style language models, diffusion, Rectified Flow, multimodal systems, plus classic discriminative models for classification, ranking, retrieval, and recommendation.
- Debug and level up existing pipelines – improve data, losses, training loops, and metrics, and make the tricky pieces legible to your engineers.
I work remotely as an embedded research engineer—either as a full-time hire or via my LLC on longer-term engagements and scoped projects.
Featured Project
Joint Image+Text Rectified Flow Training Loop
Building a joint image and text Rectified Flow model from scratch in PyTorch. This video walks through the complete training loop implementation, demonstrating the mathematical foundations and practical engineering of modern generative models.
Experience
Industry experience at scale
Machine Learning Consultant
IndependentRemote
Advising clients on ML system design and implementation. Building custom models from first principles for specialized domains and getting them into production.
Machine Learning Content Creator
YouTube Channel: @JustinTheJediRemote
Publishing state-of-the-art PyTorch implementations and first-principles math explanations of deep learning models. Topics: Rectified Flow, Continuous Normalizing Flow, Stable Diffusion XL, CLIP, GPT, Vision Transformers, ResNet, UNet, VAEs, AutoEncoderKL.
Senior Machine Learning Scientist
SpotterLos Angeles, CA
Created multi-agent LLM pipelines for content generation. Fine-tuned SDXL using LoRA and DreamBooth. Combined GPT-3.5, E5 embeddings, clustering, and LOF for uniqueness scoring.
Professional MMA Fighter
Cage WarriorsRemote
Competed professionally in mixed martial arts as an exclusive athlete for the number one MMA promotion in Europe
ML Consultant
IndependentRemote
Built predictive ML systems for UFC fights using scraped fighter data and custom feature engineering.
Machine Learning Scientist
AmazonRemote
Led Prime Video cover art personalization via intelligent image recommendations. Improved global streaming by 2.9% through ML-optimized visuals. Built distributed Bayesian models, multi-armed bandits, and topic model-based clustering pipelines.
Senior Software Engineer
CSRA IncWashington, DC
Led enterprise Java app development, Docker-based testing, and prototyping.
Technical Deep Dives
Building advanced ML architectures from first principles
NanoGPT from scratch in PyTorch
Complete implementation of a GPT-style transformer language model, covering attention mechanisms, positional encodings, and autoregressive training.
CLIP from scratch in PyTorch
Building OpenAI's CLIP model from the ground up—dual encoders for vision and language with contrastive learning objectives.
Vision Transformer from scratch in PyTorch
Building the Vision Transformer (ViT) architecture—patch embeddings, transformer blocks, and classification heads for image recognition.
Rectified Flow for Image in PyTorch: Train Loop (Part 1)
Complete PyTorch implementation of Rectified Flow for image generation—building the model architecture, training loop, and sampling process from the ground up.
Stable Diffusion XL Objective Function Derivation
Mathematical derivation of the SDXL training objective, from variational lower bounds to practical noise scheduling strategies.
Rectified Flow objective explained
Mathematical breakdown of the Rectified Flow training objective for generative models—from continuous normalizing flows to practical implementation.
AutoencoderKL from scratch in PyTorch
Implementing the KL-regularized autoencoder used in latent diffusion models—encoder, decoder, and KL divergence loss.
Deploying and training NanoGPT on Runpod
Practical guide to deploying GPT models on cloud infrastructure—setting up training pipelines, managing compute resources, and monitoring experiments.
Education
MS in Computer Science
(former PhD track)University of Central Florida
BS in Mathematics and Computer Science
Dual Major
Christopher Newport University
GEM Fellowship
National GEM Consortium