What makes Brevis remarkable? Is it Pico Prism shattering real-time proving records with 99.6% coverage of Ethereum blocks? The dozens of live integrations across top protocols like PancakeSwap, MetaMask, or Linea? Novel approaches like continuous protocol incentivization and privacy-preserving InfoFi that no one else is building?
All of it, of course. But these achievements didn’t appear out of nowhere. The real question is who made them possible?
This series explores the talent behind the tech, the remarkable individuals whose skills, knowledge, and experience comprise everything Brevis represents today.
First up: Alan Li, our Chief Science Officer.
Scientific Foundations Meet Entrepreneurial Execution
What we’re building at Brevis sits at the cutting edge of zero-knowledge technologies. There’s no established blueprint to follow and no safe path validated by years of industry practice. We’re dealing with some of the most complex mathematical science there is, in a field evolving so rapidly that expertise from a few years ago is already outdated.
This demands something specific from leadership that few have had the opportunity to develop. You need mathematical intuition to see optimization opportunities in complex systems. You need production experience to know which theoretical advances actually scale versus which just publish well. And you need the flexibility to master entirely new domains fast enough to stay ahead of a field that reinvents itself constantly.
Alan has spent his career building exactly this combination of skills.
His mathematical foundation came from MIT, where he earned his PhD in Electrical Engineering and Computer Science working on AI and mathematics. There he developed an intuition for where computation hides in complex systems and how to exploit structure to reduce it without sacrificing correctness. The kind of intuition that transfers across domains because it operates at the level of mathematical structure rather than any single application.
That foundation would prove its value in unexpected territory.
Before joining Brevis, Alan founded and led an AI for Science startup that has successfully advanced three drug pipelines to the development candidate stage and published breakthrough research in Nature and Cell. More significantly, they convinced pharmaceutical companies to commit serious capital based on AI predictions accurate enough to advance compounds through multi-million dollar development processes.
The experience equipped Alan with precisely what cutting-edge ZK demands. He entered an unfamiliar field, mastered its complexities quickly, and built production systems that companies bet capital on. When there’s no established blueprint and the technology evolves faster than traditional expertise can accumulate, that flexibility matters more than decades in any single domain.
This is what makes Alan exceptional as Brevis’s Chief Science Officer. Not just the mathematical foundations from MIT or the proven execution from building a successful company, but the combination of both, deployed with the adaptability to produce results in any technical environment. When he joined Brevis, that combination produced results almost immediately.
From Neural Networks to Zero Knowledge
Alan’s first contribution at Brevis came within months of joining, tackling one of zkML’s most stubborn problems. Verifiable machine learning has become critical as AI integrates with blockchain, but there’s a reason most zkML remains stuck in research papers rather than production systems. The computational cost of proving ML operations scales so poorly that real-world deployment stays out of reach.
Alan recognized something others missed. Neural networks are naturally sparse, with most parameters contributing minimally to outputs. Yet zkML systems were treating every operation as equally important, generating expensive proofs for computations that barely mattered. This was exactly the kind of optimization problem Alan had spent years solving, recognizing hidden structure in complex systems and exploiting it to reduce computational work.
SpaZK, the protocol he developed, generates proofs only for the operations that significantly impact outputs, leveraging the natural sparsity in neural networks to dramatically reduce computational costs. The framework is now integrated into Pico’s coprocessor architecture and already powers a partnership with Mendi Finance, where it enables ML-driven risk parameters for decentralized lending. As AI becomes essential infrastructure for blockchain applications, SpaZK gives Brevis a foundation few competitors can match.
But zkML was just where Alan started. The bigger challenge was already waiting.
Real-Time Proving: The Impossible Made Routine
Real-time Ethereum proving has been the defining challenge for zkVMs in recent years. The concept is deceptively simple: generate cryptographic proofs for Ethereum blocks faster than the network produces new ones. The Ethereum Foundation set clear targets for what they call “home proving,” thresholds that would let solo stakers run provers from home rather than requiring industrial infrastructure. Prove 99% of blocks in under 10 seconds, using hardware under $100K, with power consumption under 10kW.
When Alan joined Brevis, the state of the art sat at 40.9% of blocks proven in 10.3 seconds, requiring 160 GPUs (roughly $256K in hardware). The gap between that reality and the Foundation’s targets looked like years of work.
Under Alan’s leadership, Pico Prism achieved 99.6% real-time coverage using 64 GPUs and $128K in hardware, with an average proving time of 6.9 seconds. Getting there took under a year and many 100-hour weeks from the team. The leap came from a distributed architecture that decomposed proving work across GPU clusters while maintaining near-linear scaling, the same pattern of finding computational shortcuts in complex systems that drove SpaZK.
The achievement earned endorsements from Vitalik Buterin and Justin Drake, signaling that Alan’s work had put Brevis at the center of conversations about Ethereum’s proving future.
Built to Solve What Others Can’t
What Alan has accomplished in his time at Brevis tells only part of the story. SpaZK and Pico Prism are significant achievements, but they also reveal something about what Brevis can expect going forward. The same instinct for finding hidden efficiencies in complex systems, the same rigor in shipping production-ready solutions, will shape whatever challenges come next.
Pico Prism’s 99.6% coverage already exceeds what seemed possible a year ago, and the work is far from done. The next milestone is achieving 99% real-time proving with fewer than 16 RTX 5090 GPUs, down from today’s 64, bringing hardware requirements to the Ethereum Foundation’s home proving vision so that solo stakers can realistically participate. Alan’s team expects to reach that target soon, and continue pushing beyond it.
Then there’s ProverNet. Brevis recently launched the mainnet beta for the first-of-its-kind decentralized marketplace for ZK proof generation, infrastructure shaped by everything we learned generating over 250 million proofs across more than 30 partners. With Alan’s leadership and the technologies his team has built, we expect ProverNet to become the coordination layer for ZK proving across the ecosystem, matching applications with specialized proving capacity at a scale no single operator could achieve alone.
This is what it means to have Alan as Chief Science Officer. When Brevis takes on hard problems, we know we have someone who will find the way through.
Brevis Network© 2023 