Engineering leader who still writes code. I get energy from hard problems, whether that's distributed systems architecture, figuring out how AI fits into real workflows, or helping engineers level up.
My thing is taking complex technical problems and making them approachable. I've done this across geospatial systems, real-time data pipelines, API platforms, and now AI tooling. The common thread: good abstractions, composable systems, and always asking "how do we make this easier for the next person?"
I read a lot, research even more, and genuinely enjoy bringing people along for the ride. Most of my best ideas come from whiteboard sessions that start with "okay hear me out..."
Cold starts and runtime stability in model serving, from first principles. Checkpoint bandwidth, scale-to-zero probability, queueing collapse, KV-cache capacity, batch-dependent decode speed, nondeterminism, and tail amplification in agent loops.
A systems design problem: how do you maintain persistent state across stateless inference runtimes? This is less about AI and more about building durable architectures when your compute layer is fundamentally ephemeral.
Modern inference runtimes are stateless by design. Every request reconstructs context from scratch. This is the same pattern we solved in web services with session stores, caches, and databases. Why are we relearning it?
When you swap between different models locally, all accumulated context evaporates. Retrieval systems fetch fragments. Memory systems inject tokens. But neither maintains relational structure across runtime switches.
What if there was a persistent graph that lives alongside your local models? A sidecar architecture that maintains entity state, timeline, and retrieval policies independent of which runtime is active.
User ↓ Inference Runtime (Ollama) ↓ Context Sidecar ├── Semantic Graph ├── Entity Memory ├── Timeline State └── Retrieval Policies
The sidecar intercepts requests, enriches context from the graph, and persists new relationships back. Runtime-agnostic. Local-first. The graph survives runtime switches, updates, even complete swaps.
Vector similarity finds related chunks but doesn't encode why they're related. The graph stores relationships explicitly: causality, temporal ordering, entity connections. Any runtime can reason over structure, not just surface similarity.
This is a systems architecture thesis. Local-first compute needs persistent identity infrastructure. The runtimes become interchangeable; the context layer becomes the product.
More half-baked ideas brewing. The best ones usually start as scribbles.
Hyper-personalization as a reflection of oneself. Technology that learns who you are, not who the average user is, and mirrors your rhythms, taste, and way of thinking back to you.
Local-first memory infrastructure for Ollama. Context transfer between models, usage tracking, catch-up briefs.
A side project exploring context lineage for AI agents. Scores websites for AI readiness.
Local-first context arbitration for resolving conflicting instructions in memory stores.
Leading AI and platform strategy. Shipped Code Studio, scaled the developer platform, and somehow convinced everyone that AI tooling is actually worth the investment.
Built the core platform integrating Zelle, Direct Deposit, Micro Deposits, and Card Management. Real-time money movement at scale. Learned that highly regulated systems can still be elegant if you fight for it.
Built an internal geocoding system using TIGER, OpenStreetMap, and proprietary geospatial datasets. Powered Zillow Group location intelligence and Zestimate spatial data. Helped take Zillow international with the Canada launch.
WorldWind virtual globe SDK. Built software that helped people look at Earth from space. Peak childhood dream energy.
