Our consulting work is an extension of our research — not a separate line of business. We take on a limited number of engagements per year, in areas where our in-house expertise is directly applicable. We do not offer generic advisory services.
Most patent work in AI is done by attorneys with limited technical depth or by engineers with no patent process experience. We sit at the intersection — able to assess the genuine novelty of a method, identify the relevant prior art accurately, and communicate the technical substance in a form that survives scrutiny.
Engagements typically involve prior art research across academic literature, existing patent databases, and open-source implementations; an assessment of claim scope and defensibility; and technical documentation support for novel filings. We work alongside existing patent counsel rather than replacing it.
We do not file patents ourselves — we provide the technical foundation that makes filings stronger and more defensible.
ML codebases accumulate technical debt in ways that are distinct from general software: silent numerical errors that don't surface until late training, data pipeline bugs that corrupt model behaviour without raising exceptions, inference optimizations that introduce correctness regressions, and security vulnerabilities specific to model serving and serialization formats.
Our audits are conducted by engineers who have built the same kinds of systems from scratch — not by generalist security firms applying checklists developed for web applications. We know where the failure modes live because we have encountered them in our own work.
Engagements are scoped to specific system components or end-to-end pipeline reviews depending on client need. We provide written findings with technical detail sufficient for an engineering team to act on directly.
Cryptographic systems fail in specific, well-understood ways: incorrect implementation of sound primitives, misuse of secure algorithms in insecure compositions, side-channel exposure from timing or memory access patterns, and consensus mechanisms with edge-case vulnerabilities that only manifest under adversarial conditions.
We approach these reviews from an engineering standpoint — examining actual implementation against specification, analysing the threat model for completeness, and assessing whether the chosen cryptographic primitives are appropriate for the stated security goals. Smart contract audits cover logic correctness, reentrancy, economic attack vectors, and gas efficiency.
We do not issue certification stamps. We produce technical findings that a development team can verify, reproduce, and act on.