Circes is the operating system we wished existed when we were running our own Amazon brand. So we built it.
Circes was founded by operators, not consultants. We spent years inside Seller Central — building, scaling, and sometimes breaking — Amazon brands. We knew what was missing: a system that could see every lever at once and act on all of them.
Today, Circes runs on a live account every day. Every recommendation, every automation, every result on this site comes from that same account.
Ten years running Amazon brands, including the one I run today. Five of those years I spent inside Amazon, building catalog-risk systems and early LLM tooling. Amazon encourages employees to sell on the platform, so I was the seller and the platform operator at the same time.
I started Circes at MIT as a thesis-in-practice: could the AI systems I helped build inside Amazon be turned outward to give independent brands the operational leverage that only the largest sellers — and Amazon itself — already have? I've been running Circes live on my own brand since day one. Every recommendation it makes for a customer's account has already been tested on mine.
Three years building production agentic AI. At TSBX, he owned the MCP tooling layer for a flagship agent builder platform, cutting model inference costs by 82% in production. That same problem — AI systems that work in theory but bleed money at scale — is one Circes solves for Amazon operators every day.
Before that, he scaled mission-critical supply chain applications at Genpact, cutting service latency by 40%. The overlap with Circes isn't incidental: inventory operations, real-time bid decisions, and multi-agent coordination under production constraints are all supply chain problems dressed in Amazon clothing.
He co-authored a published ACM paper on LLM applications. He joins Circes to architect the multi-agent infrastructure and the Knowledge Ledger — the compounding data asset that makes Circes harder to replicate with every account we add.