Building AI‑native products at the intersection of capital, craft, and scale.

Product Manager on Hub Operations at DroneUp, owning the technology platform underneath a 31-hub autonomous delivery network supporting Walmart and Chick-fil-A operations. The “hub operations” framing was misleading internally — the real work wasn't running hubs, it was building the platform that hubs ran on: the autonomous orchestration layer, the fleet telemetry system, and the observability infrastructure that made any of it intelligible to the people responsible for it.

The platform underneath the network had been built quickly to get hubs operational. It worked at small scale, but it wasn't architected for what the business needed to become. Telemetry was thin and inconsistent across hubs. Observability into fleet behavior was reactive — incidents got diagnosed after the fact, often from logs reconstructed manually. The orchestration logic was monolithic, which meant changes anywhere risked breaking things everywhere. None of this was a crisis at 31 hubs. All of it was going to be a crisis at 50, and a fatal one at 100.

Two architectural calls had to be made at the CTO and SVP of Product level, and they were entangled. First: keep the monolith and extend it, or break the platform into distributed services that could scale and fail independently. The monolith was familiar and shipped fast in the short term. The distributed path was correct for the trajectory but expensive and disruptive to migrate to. Second: build telemetry and fleet observability in-house, or integrate an external optimization and observability vendor. In-house meant defensible IP, full control of the data shape, and the ability to feed it back into the orchestration layer. External meant faster time-to-coverage and lower upfront cost, but commodity capability and a dependency on someone else's roadmap for a system that was core to the business. Neither call was obviously right in isolation, and they couldn't be made independently — the architecture choice shaped which observability path made sense, and vice versa.

I structured the analysis around three lenses, designed for an audience that was equal parts technical and commercial.

First, what does the platform look like at 5x scale? I projected the system behavior of both architectural paths against a network growing from 31 to 150+ hubs. The monolith path didn't just slow down — it failed in specific, predictable ways: deployment risk, blast radius on incidents, inability to evolve subsystems independently. The distributed path had higher upfront cost but absorbed scale without architectural rework. That framing reframed the cost comparison entirely.

Second, what's the defensibility cost of outsourcing telemetry? External optimization vendors are fine for commodity problems. Autonomous fleet behavior at scale is not a commodity problem — the telemetry shape itself is competitive IP, and the feedback loops between observability and orchestration are where the network's actual intelligence lives. Outsourcing meant renting back our own data in degraded form. I made the case that in-house telemetry was a moat, not a cost center.

Third, what does the enterprise customer conversation actually require? Walmart and Chick-fil-A weren't buying drones. They were buying confidence in a system. Confidence required observability they could see, SLAs the platform could credibly defend, and a roadmap that didn't bottleneck on a third-party vendor's priorities. I made sure leadership saw the platform decisions through the lens of what the commercial relationship demanded, not just what engineering preferred.

The platform was rebuilt around distributed services and in-house telemetry. Fleet behavior became observable in resolution that wasn't possible before — incidents shifted from post-hoc reconstruction to real-time visibility, and orchestration decisions started using telemetry as an input rather than a record. New hub launches stopped requiring proportional engineering effort. The platform absorbed scale instead of fighting it. More importantly, the architecture decisions are still the foundation the network runs on — they were correct against a trajectory that has continued to play out exactly as projected.

Product Partnerships at hireEZ, an AI-driven recruiting and sourcing platform. The platform was strong at generalist sourcing but couldn't credibly serve healthcare — a high-value vertical with specialized credentialing, licensing, and matching requirements that public web data alone couldn't support.

Healthcare was the obvious next vertical for hireEZ — high recruiter spend, persistent talent shortages, willingness to pay for specialized tooling. But the platform's core data layer was built on public web sources, which couldn't reliably surface license status, credential verification, or specialty-specific matching signals. Without that data, the product couldn't credibly compete against healthcare-native recruiting tools.

Three real options. Build the healthcare data layer internally — slow, expensive, uncertain to converge on reliable data. License it from existing data providers — fast, but commodity, no defensibility, and the unit economics didn't work at scale. Or negotiate proprietary partnerships directly with credentialing and licensing data sources — the hardest path, the longest sales cycles, and the only one that produced a durable moat if it worked.

Two lenses drove the decision.

First, what's the smallest set of partnerships that actually unlocks the use case? Not the maximalist version. The team's instinct was to map every possible healthcare data source and pursue all of them. I argued the opposite — find the two or three data partnerships that, combined, would let us credibly say “hireEZ does healthcare” to a buyer. Anything beyond that was scope creep dressed as thoroughness.

Second, what's the cost of being wrong about defensibility? If competitors got the same data six months later through the same providers, the partnership strategy was just a delayed commodity. So the partnerships had to include some structural exclusivity, time-window advantage, or data shape that couldn't be replicated by signing the same vendor. That requirement narrowed the negotiation playbook significantly and made the conversations harder, but it changed what we were actually building.

Negotiated the proprietary partnerships that became the foundation for hireEZ's healthcare offering — done before AI was the default narrative in recruiting. The product line unlocked an estimated $10M ARR revenue opportunity and a vertical the company hadn't previously been able to credibly enter. The partnership structure also became the template for how hireEZ approached subsequent vertical expansion.