If you want to change the world, don't protest. Write code.
I've been writing production software for 15+ years. Payments, banking, blockchain, cryptography. Mostly in situations where getting it wrong costs real money or breaks real trust.
Head of Ecosystem at StarkWare. Co-authored EIP-1559. Started Kakarot and Madara from nothing, both now run by independent teams. Lately I spend most of my time thinking about AI safety and what happens when autonomous systems need to prove they did what they said they did.
AI is about to be everywhere and in everything. I don't think we've thought nearly enough about how to verify what it's actually doing. That's what I work on now: cryptographic proofs and governance tooling to keep AI systems honest.
| Project | What it does |
|---|---|
| awesome-ai-safety | Curated list of tools and resources for AI safety. Alignment, interpretability, red teaming, formal verification, ZKML, governance. Focused on things you can actually use, not just papers. |
| eu-ai-act-toolkit | Open source toolkit for EU AI Act compliance. SDK, CLI, and web app for classifying AI systems and generating compliance docs. |
| llm-provable-computer | Can you prove an LLM produced a specific output without running it again? Exploring verifiable inference with STARKs. |
I also write about this:
| Math Is Humanity's Last Bastion Against Skynet | Why ZK proofs are the foundation for AI safety at scale |
| Can LLMs Be Provable Computers? | Verifiable AI inference via STARKs |
I implement frontier ML papers in Rust. It's how I learn, if you can build it from scratch, you understand it.
| Project | Paper / architecture |
|---|---|
| jepa-rs | First Rust implementation of JEPA primitives (I-JEPA, V-JEPA, C-JEPA, VICReg, EMA) |
| gpc_rs | Generative robot policies. Diffusion policy + world model + evaluator, in Rust. |
| mosaicmem | Geometry-aware spatial memory for camera-controlled video generation |
| attnres | Attention Residuals (Kimi/MoonshotAI). Softmax attention over all preceding layer outputs. |
| turboquant | Google's TurboQuant. Vector quantization of LLM KV caches, in Rust. |
| jepa-notebooks | Interactive notebooks exploring JEPA architectures |
Co-authored EIP-1559, the fee market reform. Shipped to mainnet August 2021.
Built the open source ecosystem around Starknet. Started Kakarot and Madara, both now have their own teams and communities.
| Project | What it does |
|---|---|
| Kakarot | EVM interpreter written in Cairo. Ethereum compatibility on Starknet via ZK proofs. |
| Madara | Starknet sequencer for sovereign appchains |
| Raito | Bitcoin ZK client in Cairo. Verifies Bitcoin consensus inside a STARK proof. |
| Askeladd | Verifiable computation for Nostr Data Vending Machines via STARKs |
| Cashu ZK Engine | Blind Diffie-Hellmann Key Exchange in Cairo for Cashu ecash |
| Project | What it does |
|---|---|
| bitcoin-mcp | Bitcoin & Lightning Network MCP Server |
| nostringer-rs | Ring signatures (SAG, BLSAG) for Nostr, in Rust |
| nostr-mcp | Nostr MCP server |
| bitcoin-honeybadger | Bitcoin Honeybadger |
| Before Fighting Banks, Let's Understand How They Actually Work | A cypherpunk's guide to the financial system |
| Time to Take the Nostr Pill | Why Nostr matters for freedom of speech |
| Nostr DVMs Meet Verifiable Computation | STARKs powering trustless Nostr services |
| Cashu Meets STARKs | Zero-knowledge proofs for the Cashu ecash protocol |
If there's a thread, it's this: the gap between what individuals can verify and what institutions can hide should be closed by math, not by trusting people to behave. Bitcoin showed money doesn't need banks. Ethereum showed computation doesn't need servers. ZK proofs are how we'll know an AI actually did what it claimed.
Billions of autonomous agents are coming. Hospitals, roads, financial systems. The trust model we have assumes a small number of known actors. That assumption is about to break.
Math scales. Goodwill doesn't.