We build and ship AI for drug discovery and healthcare.
Move37 Labs is an AI consultancy founded by a former Google AI leader and biotech executive. We partner with tech bio, medtech, and impact-driven organizations to turn cutting-edge research into products, publications, and strategy.
Trusted by researchers and teams at
Joel Shor
Former Google AI leader and biotech VP who has shipped production AI systems used by millions, and has published across NeurIPS, ICML, Nature, and CVPR.
What We Do
AI Strategy
Identify high-value AI opportunities in your pipeline, from target identification to clinical workflows. We build tailored project plans based on deep diligence and competitive analysis.
Fractional AI Leadership
Embed senior AI leadership to keep technical teams focused and translate progress for the C-suite. We identify blockers, mentor researchers, and guide critical pivot-or-persist decisions.
Custom AI Models
Select and adapt the right open-source or foundation models for your specific biological data. We fine-tune, evaluate, and deploy purpose-built models for your therapeutic area.
Data Architecture
Design scalable data infrastructure and management systems that grow with your organization. We guide build-vs-buy decisions and implement engineering best practices for long-term maintainability.
Research & Publishing
Translate your AI achievements into high-impact publications that attract funding and talent. We handle study planning, rigorous evaluation, and narrative crafting from preprint to peer review.
Technical Due Diligence
Evaluate AI capabilities, technical debt, and team strength for M&A, investment, or partnership decisions. We provide the technical depth that deal teams need.
Inference Optimization
Reduce production model cost and latency with expert inference optimization on GPUs and TPUs. We leverage hands-on experience and partnerships with NVIDIA and Google Cloud.
Selected Work
State-of-the-Art Nucleic Acid Sequence Design
Led a research collaboration between Google, MIT, and Move37 Labs to build the first large-scale benchmark for modern neural nucleic acid design algorithms, and developed AdaBeam, a state-of-the-art model-based optimization designer that outperformed existing approaches.
Outcomes
- Accepted to the 2025 ICML Workshop on Generative Biology
- In review at Nature Machine Intelligence
- Benchmark and designer open-sourced
- Featured on the Google AI Research Blog
"Joel led the NucleoBench collaboration across Google, MIT, and Move37 Labs with a rare combination of scientific depth and project leadership. He drove the benchmark design, developed AdaBeam into a state-of-the-art designer, and kept a complex multi-institution effort focused and productive. He's one of the few people I've worked with who can operate at the frontier of ML research and simultaneously manage the execution to get it published and shipped."
Structure Prediction for Drug Discovery
Designed model infrastructure and fine-tuned best-in-class structure prediction and chemprop models on the company's proprietary data. The resulting model zoo is used by bench scientists to accelerate lead optimization in an active drug discovery pipeline.
Outcomes
- Models deployed into production drug discovery workflow
- Used by bench scientists to accelerate lead optimization
- Built reusable infrastructure for ongoing model development
Other Engagements
Genomic Language Models
University of Texas, AustinLeading research into novel training approaches for genomic language models with UT Austin graduate students.
Genomic Language Models for Marine Biodiversity
New Atlantis LabsFractional Head of Data Science. Developed gLM-based taxonomic classification of marine environmental DNA, including classification of truly unseen species.
AI Agents for Healthcare App Generation
QuomeLed development of a benchmark for evaluating AI agent-driven healthcare app generation. Published at AgentCraft Workshop, ACM IUI 2026.
Fetal Heart Abnormality Detection
HeartSoundsBuilt a product to detect fetal heart abnormalities from at-home ambulatory doppler ultrasound. Secured provisional patents and applied for government funding.
Benchmarking LLMs on Mechanical Engineering
Boston UniversityCollaborated with MechE lab to create and publish FEM-Bench, a benchmark of LLM performance on mechanical engineering reasoning tasks.
Vocal Biomarkers for Perinatal Depression
NurtrContributed to algorithm development for detecting perinatal depression from voice biomarkers. Published on arXiv.
Biotech Competitive Intelligence
Biotech intelligence startupHelped architect and benchmark an agentic system for producing automated competitive landscape reports for biopharma companies.
Targeted Drug Design
Early-stage biotech startupTechnical advisor on applications of ML to therapeutic design.
Protein Structure Prediction
Stealth startupImproved inference speed and cost on next-generation protein structure prediction models.
Experiment Design Infrastructure
Flagship portfolio companyHelped design experiment design infrastructure using Bayesian Optimization.
FDA Submission Generation
Biopharma clientBuilt a production LLM system for generating FDA regulatory submission documents.
Earlier Career Highlights
| Project | Impact |
|---|---|
| Next-Generation TPUs for Genomic Foundation Models | As VP & Head of ML, led a collaboration with Google Cloud to run RNA target identification models across the entire human genome using latest-generation TPU hardware. Enabled genome-wide ML inference at a scale previously impractical. Published on the Google Cloud Blog. |
| TensorFlow-GAN | Designed and authored Google's open-source standard library for training and evaluating GANs. 6.5M+ downloads, patented, associated course taken by 200,000+ students. Papers built on TF-GAN have accumulated thousands of citations. |
| Project Euphonia — AI for Speech Accessibility | Founded the team that pioneered AI-powered speech accessibility for people with ALS and other conditions. Announced at Google I/O, featured in a documentary narrated by Robert Downey Jr., and catalyzed a Google/Amazon/Microsoft collaboration. |
| Japan's National COVID-19 Forecasting Initiative | Led development of 28-day public forecasts used by the Japanese government to shape pandemic policy. Seen by millions. Published in npj Digital Medicine. Won the VLED Award for "Enhancing Public Good with Big Data." |
| Real-Time Polyp Detection for Colonoscopy (Verily) | Developed AI-powered real-time polyp detection during colonoscopies. Received approximately $100M acquisition offer from a major endoscope manufacturer. |
| First Human-Indistinguishable Text-to-Speech (Tacotron) | Co-developed the TTS system and style control techniques that form the foundation of modern text-to-speech. Papers have 1,000+ citations each. |
| BigSSL — Large-Scale Semi-Supervised Speech Model | Co-built a 600M+ parameter semi-supervised speech model. Published in IEEE JSTSP, 254 citations. |
| DNA Sequence Error Correction (Verily & Google) | Knowledge-distilled a transformer-based error-correction model for long-read DNA sequencing. Runs on-device on PacBio sequencing machines. Published at NeurIPS workshop. |
Publications & Patents
FEM-Bench: A Structured Scientific Reasoning Benchmark for Evaluating Code-Generating LLMs
From Prompt to Product: A Human-Centered Benchmark of Agentic App Generation Systems
Voice biomarkers of perinatal depression
A prospective evaluation of AI-augmented epidemiology to forecast COVID-19 in the USA and Japan
Knowledge distillation for fast and accurate DNA sequence correction
The Need for Medically Aware Video Compression in Gastroenterology
Does video compression affect CADe polyp detectors?
Predicting the generalization of CADe models for colonoscopy
Reducing health anxiety in IBD patients using video testimonials
Combined Compression and Feature Extraction Models for Medical Videos
How We Work
Flexible engagement models tailored to your stage and needs.
Insights & Intel
Stay connected to cutting-edge developments in biotech AI.
- Biotech AI newsletter with industry analysis
- Free, unsubscribe any time
Advisory Engagement
One-on-one sessions with a senior biotech AI researcher and executive.
- Specific technical questions & architecture reviews
- Strategic decisions & second opinions
- Flexible scheduling, virtual or in-person
Retained Partnership
Ongoing AI leadership and execution for select organizations.
- Custom model development
- Strategy implementation & team building
- Limited availability
Joel Shor
Founder, Move37 Labs
Joel Shor is the founder of Move37 Labs, an AI consultancy focused on drug discovery, genomics, and healthcare. He brings over a decade of experience building and leading AI teams at Google, Alphabet's Verily, and Deep Genomics — shipping production systems that have reached millions of users, publishing at top venues, and leading organizations from research through deployment.
At Google, Joel spent eight years as a Staff Research Scientist and founded two teams: the Audio Biomarkers team, which pioneered Project Euphonia (AI-powered speech accessibility for people with ALS, announced at Google I/O and featured in a documentary narrated by Robert Downey Jr.), and AI for Japan, which built the national COVID-19 forecasting system used by government policymakers and seen by millions. His research contributions include TensorFlow-GAN (6.5M+ downloads, Google's standard GAN library), co-development of the first human-indistinguishable text-to-speech system (Tacotron), and foundational work in speech representation learning.
Joel then served as VP and Head of Machine Learning at Deep Genomics, a Series C biotech startup with over $200M in funding, where he directed cross-functional teams of AI researchers and engineers, led collaborations with Google and NVIDIA, and managed large-scale compute infrastructure for drug discovery. He is currently a Visiting Researcher at the University of Texas at Austin and Boston University, and a speaker at events including Foresight Institute's Vision Weekend.
Joel's work has been published in Nature, NeurIPS, ICML, CVPR, Interspeech, and npj Digital Medicine, among other venues. He holds 12 patents, an i10-index of 19, and multiple papers with over 1,000 citations. He is the sole recipient of Princeton's Manfred Pyka Memorial Prize for outstanding work in physics.
"Working with Joel fundamentally changed how I approach AI research. He has a gift for breaking complex problems into clear, tractable pieces and knowing which ones matter most. He pushed me to think more critically about evaluation methodology and experimental design, and the work we published together is something I'm genuinely proud of. Any team that gets to work with Joel is going to level up."
Let's build your AI strategy together.
Whether you need a one-time advisory session, ongoing AI leadership, or help turning your research into a publication, we'd like to hear from you.