I design and build AI systems that have to work under real constraints.

Private deployment, tight hardware budgets, agents that must behave, and participants with a reason to cheat. As a principal systems architect, I come in before the architecture is obvious: I define the problem, build the critical path, and make the risky assumptions testable while they're still cheap to change.

Discuss an engagement See the problems I work on Download résumé
Operating envelope 1 / 3
Real-time underwater detection
latency
<50 ms end to end
hardware
Jetson, embedded
input
operational noise
training
simulation to real
tuned for
Pd against Pfa
  1. DSP features
  2. Deep CNN
  3. operator
binding constraint

The work starts before the architecture is obvious.

In these problems the constraints are real and the obvious abstraction is usually wrong. The highest-leverage engagements are usually not cleanly specified engineering tasks. They are systems problems where several constraints interact and the team needs a sharper model of the problem before adding more implementation.

  1. 1

    Define the system

    Turn fragmented requirements, competing assumptions, operating constraints, and failure modes into a coherent architecture and executable roadmap.

  2. 2

    Build the critical path

    Prototype and implement the load-bearing components until the architecture has been tested against real software, hardware, data, users, and operational conditions.

  3. 3

    De-risk the system

    Make latency, security, reliability, model behavior, cost, adversarial incentives, deployment, and recovery explicit before they become expensive production surprises.

Problems I'm brought in for

ProblemWhere I've solved itOutcome
Your AI has to run privately or offline Secure distributed ML Adopted for ongoing production use Inference has to fit a hardware and latency budget Real-time underwater detection Sustained sub-50 ms inference on embedded hardware Your agent workflow has to behave, across many customers TurnkeyHQ Multi-tenant AI platform in private beta Participants have a reason to game the system Adversarial storage Operated in production under continuous integrity proofs
All systems work and case studies

When am I most useful?

Usually when the problem sounds something like this: “We know what outcome we need, but several systems, constraints, and stakeholders interact; the existing abstractions are breaking down; and simply adding more engineers will not tell us what the architecture should be.”

That is a decision problem, not a staffing problem. Teams in it are usually better served by a bounded review of the decision than by committing another engineer for six months to a design nobody has attacked yet.

Bring the problem, the constraints, and the desired outcome.

See engagement types

Know someone working on one of these problems? Introductions are welcome.

Independent research & labs

Alongside production systems work, I maintain longer-horizon research programs in persistent machine cognition, verifiable research infrastructure, monetary architecture, optimization, and mathematical form.

Ask