We are specifically interested in the intersection of hardware and software — how architectural choices at the silicon level shape what AI systems can and cannot do in production, and how to design models that perform where it matters rather than where benchmarks are favorable. In deep tech, runtime and precision must both be satisfied. That constraint is not a limitation we work around; it is the design criterion we start with.
We are also investing in AI as a strategic participant in discovery — systems capable of abductive reasoning, of forming and testing hypotheses under genuine uncertainty, and of operating as genuine collaborators rather than passive tools. The goal is AI that advances the science, not merely accelerates the paperwork.
TAP's AI team spans foundational work in machine learning — across architecture design, large-scale systems, and reasoning under uncertainty. Members of this team have contributed to the development of AI infrastructure, not just its application, and bring a grounded view of where current capabilities hold up and where they fall short in practice.
The gap between what AI can do in theory and what it can do in production is a hardware-software problem. TAP's team works at this boundary — designing architectures that account for memory constraints, latency requirements, and computational cost from the outset. This matters especially in deep tech, where deployment environments are specialized and unforgiving.
In deep tech, AI does not get to choose easy problems. Runtime and precision must both be satisfied — simultaneously, under regulatory and operational constraints that cannot be relaxed. TAP's approach is shaped by this reality: building systems that hold up where benchmarks do not matter and real-world performance does.
Most AI excels at interpolation within well-defined distributions. Frontier discovery is different: data is sparse, mechanisms are incompletely understood, and the most important questions are the ones no model can yet answer confidently. TAP invests in AI systems that reason abductively — forming and stress-testing hypotheses under genuine uncertainty, supported by expert human judgment at every step.
The highest-value application of AI in deep tech is not automation — it is participation. TAP builds AI systems that surface non-obvious connections, propose hypotheses, challenge assumptions, and operate as genuine collaborators in research and decision-making. We are interested in AI that advances the science, not merely accelerates the paperwork.
Machine Learning, Deep Learning • AI Researcher/Instructor, Berkeley, NASA • Founder, Silicon Valley Bank AI Lab
Machine Learning, Computational Biology, Cancer Genomics • CSO / CTO, Cancer IQ
Quantum Computing • Senior Engineer, IonQ • Machine Learning & Physics • Financial Industry Quant
Corporate and Technology Strategy • Entrepreneur, Co-Founder ValueAI Institute • Investor, Venture Partner
Principal Scientist, O'Donnell Data Science & Research Computing Institute, SMU • HPC Applications Scientist, SMU • Adjunct Professor of Data Science, SMU • Computational Chemistry & Molecular Dynamics
Electrical Engineer • Sr TPM, WhatsApp (Meta) & Amazon Alexa AI • Founder, RoutineHub (50K → 1.4M users) • Thinks across physical systems, software, and scale • USC MBA
Head of AI, RealAvatar (Andrew Ng's AI Fund) • ML Science Manager, Amazon • AI Researcher, Megagon Labs • Lecturer, UC Berkeley MIDS • PhD Computer Science, University of Chicago
Founder, Ember Agentic Labs • Analytics & Data Science Engineering Manager, Google Cloud • Architected Google Cloud's first unified revenue data platform supporting 8,000+ daily users • Berkeley MIDS