About
What I’m building
Avarieux is a multi-source AI research platform for self-directed investors and registered investment advisors. The architectural premise is one I haven’t seen applied rigorously anywhere else: AI used in regulated domains has to be structurally honest, not just usually right. Every numeric claim is audited against the underlying source before it leaves the system. Unverifiable claims are flagged, never silently passed. Every analysis is archived as a permanent, timestamped, citable URL — so the work you do in Avarieux is reproducible by anyone, including a compliance officer.
The system operates under §202(a)(11)(D) of the Investment Advisers Act of 1940 — the publisher exclusion that the Wall Street Journal occupies. Avarieux is structurally non-advisory by design, not by disclaimer.
Incorporated in Delaware as a C-corporation via Stripe Atlas, May 7, 2026. Publicly announced May 20, 2026. Waitlist and early access at avarieux.com.
How I got here
I came out of NJIT’s data science program in late 2025 with a specific problem I couldn’t stop thinking about: the models everyone was deploying into financial workflows had no citation layer. They produced fluent, confident text that was unauditable at the claim level. In a domain where a single misquoted number can constitute a material misstatement, that seemed like a design flaw worth fixing.
While I was working through that problem, I spent about a year building in the MCP ecosystem — authoring four open-source servers, contributing patches back to Anthropic’s official repository, and developing a clear picture of what the tool layer for agentic AI actually needs to look like. That infrastructure work is what eventually became the technical foundation of Avarieux.
I’m also a Founding Engineer at Papex, a NYC fintech building receipt intelligence for Indian freelancers. That work has kept me close to the practical engineering of financial data systems while Avarieux moves toward launch.
The work I keep returning to
MCP architecture. The Model Context Protocol is the most interesting infrastructure layer to emerge in the agentic AI space — it’s the standard that lets language models actually do things rather than just describe them. I’ve written four servers, contributed to the spec implementation, and spoken publicly about what the tool layer needs to look like for agents operating in high-stakes domains.
Verification in regulated domains. The deeper question underneath Avarieux: how do you build an AI system that is trustworthy not because it’s usually accurate but because its architecture makes accuracy auditable? That’s a systems design problem, a legal design problem, and a product problem simultaneously.
The gap between what AI can do and what it can be held accountable for. That’s the territory I work in.