Jeffrey Urban

About Jeffrey Urban

I build deep learning platforms for physical systems. My work follows a pattern: identify hard problems early, design novel solutions, build from concept to production, prove it scales, then move to the next challenge.I\'ve held end-to-end ownership of full data lifecycles: innovated data sources, architected pipelines, built data audit workflows. I\'ve led projects from concept to production: defined problem frames, delivered prototypes, built and operated device and cloud-service programs at scale. I\'ve researched, designed and shipped novel algorithms as production services.What draws me to problems: complexity from physical uncertainty, distributed system dynamics, or resource constraints at the edge. I prefer challenging problems on unique data with ownership of modeling and evaluation. I build developer tools that enable this work.Over 13 years, I\'ve uncovered insights from latent data flows in connected hardware-software platforms across automotive, energy, and safety-critical systems. I turn noisy sensor data and telemetry into robust systems through dataset curation, experiment design, and evaluation frameworks.I\'m among the first 10 to complete a Master of Science in AI from UT Austin (3.9 GPA), with coursework in deep learning, reinforcement learning, NLP, and optimization implemented in PyTorch, and teaching experience in generative modeling. I work with open models (CNNs, LLMs, VLMs).I run Banche Labs, a personal studio exploring deep learning and generative AI applications.If you\'re building something ambitious where recipes don\'t exist, let\'s talk.

Experience

1. Deep Learning Engineer & Ai Agent Team Tech Lead

   Banche Labs, Inc

   Apr 2019 — Present

   Built learned-prior guided text capture system from video, combining heuristics, computer vision, and tuned vision-language models. Custom architecture samples video context to identify text positions, pinpoint subtitle transitions, and disambiguate characters using surrounding context. Low-touch annotation pipeline efficiently produced 50k validated human-AI collaborative annotations, balancing automation with strategic human validation.Also working on language fluency modeling for teaching and evaluation, extracting and recombining parts of foundation models to build bespoke models for niche applications, and building collaborative workflows with a team of AI agents working simultaneously for software development. Establishing friendships with the agents before they take over completely.Banche Labs is my personal studio for exploring applications of deep learning and generative AI.

Education

• Harvard Extension School

  Graduate Studies, Computer Science

• The University of Texas at Austin

  Master of Science, Artificial Intelligence

• University of Connecticut

  BSE, Computer Engineering

Skills

• Product Development
• Perl
• Embedded Systems
• Opengl
• Real-Time Operating Systems (Rtos)
• Threadx
• Product Strategy
• Automotive
• C
• Programming
• Quality Assurance
• Labview
• Competitive Analysis
• White Papers
• Nuclear
• Integration
• Engineering
• Renewable Energy
• Wiced Sdk
• Shell Scripting
• Strategy
• Sql
• Linux
• Swift
• Matlab
• Android
• Custom Test Systems
• Electronics
• Iot
• Qnx
• Test Automation
• Network Administration
• Product Management
• Hardware Security
• Field-Programmable Gate Arrays (Fpga)
• Electronics Rework
• Operating Systems
• Bench-Top Equipment
• Algorithms
• Light Measurement Tools