How I work
I'm a product manager working across two frontiers, crypto and AI. What pulls me in is the new primitive each one is producing. In crypto it is the vault, a programmable home for capital. In AI it is the agent, software that acts on its own. New primitives are where the interesting product work lives. I came up through ecosystem and grants work in Web3, and today I ship institutional crypto products and applied-AI systems.
On crypto I go deep on what makes a vault safe: staking and validator economics, leverage and liquidation math, the carry that pays or inverts, tokenomics, and on-chain data I query myself. On AI I build and ship: RAG over real data, fine-tuned open models, and autonomous agents that transact on their own rails. The connective tissue is judgment about what to build, and proof that it works.
Decisions settled by on-chain data, not decks
When I claim a leverage is the sweet spot, or that a strategy survives a depeg, it is read off a live query anyone can open and re-run. I write the SQL myself, on Dune, against real on-chain state. Opinions are cheap. A reproducible number is not, and it is what gets a risk committee to a yes.
Deep on DeFi mechanics, not just the narrative
Staking and validator economics, the looping math behind leveraged staking, loan-to-value and liquidation thresholds, how carry inverts when borrow demand spikes, tokenomics and governance. I work at the level where the product actually lives or dies, then package it so a fund can hold it.
I ship applied AI, I do not just prototype it
Retrieval over real databases with retrieval scored separately from generation, and fine-tuning of open models (mostly Gemma 4 on a single A100 with LoRA/QLoRA) when prompting and RAG run out of road. The unglamorous middle, grounding and specialization, is what decides whether an AI product is trustworthy or just a good demo.
Agents are economic actors now
When an agent can act, retrieve and pay per request over a rail like x402, the work moves from intelligence to control. I scope autonomy to the blast radius, treat every autonomous payment as irreversible, and wrap it in hard guardrails: spend caps, allowlists, approval thresholds, an audit trail. The upside is a genuinely new market. The discipline is what makes it shippable.
For AI, the eval is the product
A model that is confidently wrong is worse than no model. So I build the eval suite before the feature, cover happy path, edge cases and adversarial inputs, score the whole agent trace rather than just the final answer, and treat passing it as the price of shipping.
Compliance carried inside, not bolted on
In crypto, the regulatory frame (MiCA in Europe) is a design input from the first sketch. Carry it early and it stops being friction. Nine times out of ten it is the thing that lets a product reach an institution at all.