—
Neel Somani isn’t building another Layer 2 anymore. The founder who raised $50 million for Eclipse, Ethereum’s fastest platform powered by the Solana Virtual Machine, has shifted his focus entirely to AI research. Specifically, to a problem most founders ignore: making machine learning systems explainable.
“Performance metrics have become a distraction,” Somani says. “Everyone’s chasing benchmarks, but no one can tell you why their model made a specific decision. That’s a massive problem when these systems are being deployed in healthcare, finance, and criminal justice.”
It’s an unusual pivot. Most Web3 founders either double down on blockchain infrastructure or join a different field in an entry-level role. Somani is doing neither. He’s applying the same systems-level thinking that helped him scale Eclipse to one of the thorniest challenges in artificial intelligence: interpretability.
From Quantitative Research to Blockchain Infrastructure
Somani’s career has always centered on solving complex systems problems. Before founding Eclipse, he worked as a quantitative researcher at Citadel’s commodities group, where he built models for global markets. That analytical foundation, combined with a triple major in computer science, mathematics, and business administration from UC Berkeley, gave him the technical depth to tackle infrastructure-level challenges.
At Berkeley, he contributed to research in type systems, differential privacy, and scalable machine learning frameworks. Those aren’t typical blockchain founder credentials. But they explain why Eclipse focused on performance and scalability rather than speculative token economics.
“The quantitative research background taught me to question assumptions,” Somani explains. “In commodities trading, you can’t just trust a model’s output. You need to understand the underlying mechanics. That same principle applies to AI systems, but most of the industry hasn’t caught up yet.”
Why Interpretability Is the Next Frontier
The shift from blockchain to AI research wasn’t random. Somani sees interpretability as a systemic risk that the industry is underestimating.
“Right now, we’re deploying models that make decisions affecting millions of people, and we can’t explain those decisions beyond vague probability scores,” he says. “That’s not just a technical problem. It’s an ethical and regulatory liability.”
His argument is grounded in practical risk assessment. Financial institutions are already facing regulatory pressure to explain automated trading decisions. Healthcare systems are being sued over algorithmic bias in diagnostic tools. Criminal justice systems are abandoning risk assessment algorithms because they can’t justify their outputs in court.
“Interpretability isn’t about making AI ‘nicer,'” Somani clarifies. “It’s about making it usable in high-stakes environments where accountability matters. You can’t deploy a black box in contexts where people’s lives are affected.”
Lessons from Scaling a High-Growth Startup
Building Eclipse taught Somani how to move fast without breaking foundational assumptions. Series A funding doesn’t guarantee product-market fit, and industry recognition doesn’t solve technical debt.
“The hardest part of scaling Eclipse wasn’t the technology,” he admits. “It was maintaining clarity about what problem we were actually solving. It’s easy to get distracted by market hype or competitor moves. The discipline is staying focused on the core value proposition.”
That discipline is now shaping his approach to AI research. Rather than chasing the latest transformer architecture or scaling laws, Somani is focused on a narrower question: how can machine learning models generate human-understandable explanations without sacrificing performance?
“There’s a false trade-off people assume between accuracy and interpretability,” he says. “The assumption is that explainable models are inherently weaker. But that’s only true if you’re using outdated methods. There’s a lot of unexplored design space.”
Supporting the Next Generation of Innovators
Beyond his research work, Somani runs a personal scholarship program supporting students in STEM fields. It’s not a typical founder move, but it reflects a broader commitment to education and talent development.
“I benefited from incredible mentors and educational opportunities at Berkeley,” he says. “The scholarship program is about creating similar pathways for students who have the potential but not the financial access.”
The program focuses on underrepresented students pursuing technical degrees, with an emphasis on research opportunities and hands-on project work. It’s a small-scale initiative, but Somani sees it as part of a longer-term investment in the innovation ecosystem.
“Technology doesn’t advance because of individual genius,” he explains. “It advances because smart people get access to resources, mentorship, and hard problems to work on. If we’re serious about building better AI systems, we need to invest in the people who will build them.”
Navigating Industry Transitions
The transition from Web3 to AI hasn’t been without challenges. Blockchain and artificial intelligence attract different communities, different investors, and different risk profiles. Moving between them requires rebuilding credibility from scratch.
“There’s definitely skepticism when someone pivots from crypto to AI,” Somani acknowledges. “People assume you’re chasing the next hype cycle. But the underlying skills transfer more than you’d think. Both fields require deep technical expertise, systems thinking, and the ability to work on problems that don’t have obvious solutions yet.”
He’s also candid about the reputational risks of working in emerging technologies. Public scrutiny comes with the territory, especially for founders who’ve raised significant capital or worked in controversial sectors.
“Resilience isn’t about ignoring criticism,” he says. “It’s about separating legitimate feedback from noise. If someone points out a real flaw in your approach, that’s valuable. If they’re just reacting to headlines, that’s less useful. The hard part is knowing the difference.”
What Comes Next
Somani is deliberately vague about timelines and commercialization plans. Unlike Eclipse, which had clear go-to-market milestones, his current research work doesn’t fit neatly into a product roadmap.
“I’m not trying to build a company right now,” he says. “I’m trying to solve a specific technical problem. If that leads to something commercial later, great. But the focus is on getting the research right first.”
That approach may frustrate investors looking for the next unicorn, but it reflects a shift in priorities. After raising $50 million and scaling a high-growth startup, Somani is more interested in foundational problems than fundraising narratives.
“There’s a version of this where I raise another round, hire a team, and build an interpretability platform,” he admits. “But that’s not the most effective path right now. The bottleneck isn’t resources. It’s clarity about what actually works.”
For now, that means staying focused on research, publishing findings, and contributing to the broader AI safety conversation. It’s a slower path than the blockchain world, but Somani seems comfortable with the trade-off.
“The Web3 space moves incredibly fast,” he reflects. “Sometimes that’s productive. Sometimes it just creates churn. AI research requires a different tempo. You need time to test ideas, run experiments, and validate assumptions. If I’ve learned anything from Eclipse, it’s that speed without direction doesn’t get you anywhere.”
—
