Speaking + podcasts
Topics I'm happy to talk about, plus the kind of audience the topic actually serves. If you're organizing a podcast, panel, or conference and the fit looks right, get in touch.
Most AI systems make claims they can't formally verify. SUM's PROOF_BOUNDARY discipline (proved vs empirical-benchmark vs shipped) is one practical pattern for separating what's auditable from what's measured. I can talk about how to carry that pattern across a portfolio of products without losing speed.
Audience fit: AI infra teams, eval+observability vendors, regulated-AI buyers.
What it takes to build an agent observability backend on OpenTelemetry GenAI + OpenInference instead of a proprietary wire format — plus the case for MCP as the read surface and a single-schema closed eval-trace loop.
Audience fit: ML platform leads, agent infrastructure communities, OTel SIG.
Why byte-identical Ed25519 signatures over JCS-canonical bytes across Python, Node, and the browser is a non-trivial property — and what it unlocks for downstream verifiers who can't trust the issuer's runtime.
Audience fit: security and trust engineering, decentralized-identity adjacent, content-provenance projects.
The practical version: how to operate a public portfolio across a dozen+ projects without overclaiming, with discipline that survives review by adversarial buyers (compliance officers, due-diligence teams). Concrete patterns; the /proof page on this site is a live example.
Audience fit: indie operators, technical-founder podcasts, consulting communities.
What it actually looks like to ship a portfolio of production AI systems as a single operator coordinated with multiple Claude/Codex instances. The good, the bad, and the discipline that keeps it honest.
Audience fit: indie hacker / one-person-business communities, dev-tool vendors.
For pitches, drop a note via the contact page with the audience, format, and rough date.