Jeremy
Suh

The Algorithm That Ate Its Own Creator

Browser Buddy had a beautifully specific mission: find the best essays and blogs on the internet and surface them for people who actually wanted to read them. Not viral. Not trending. The long-tail stuff - the Substack writers, the personal blogs, the thinking that never makes it to the front page of anything. Jeremy Suh and Arnav Wadehra built it through Y Combinator's Winter 2024 batch and raised $500K. It worked well enough that Exa - the search infrastructure company rewriting how AI retrieves information - made it their first acquisition ever.

Will Bryk, Exa's CEO, framed the acquisition in a way that mattered: he'd known Jeremy and Arnav for two years before buying their company. They were among Exa's earliest customers. When Bryk called them "two of the smartest and kindest engineers" he'd ever met, that wasn't press release language - it was the closing of a circle. These were people who had been using Exa's tools to build something, figured out new approaches to retrieval in the process, and ended up joining to apply those insights at a much larger scale.

That phrase - "all the world's information" - is doing significant work. Browser Buddy indexed a slice of the internet. Exa indexes over 500 billion URLs and serves 400,000+ developers. The problem Jeremy is now working on is the same problem, scaled to something approaching totality.

What Exa Actually Does (And Why It Matters Now)

The standard search engine was built for humans clicking links. The search engine Jeremy now works within was built for AI systems that need to retrieve facts, verify claims, and ground responses in real, current information. Exa's API lets developers build RAG pipelines, research tools, and AI applications that can actually look things up - not hallucinate them.

The technical backbone involves neural search, embeddings, semantic retrieval, and an infrastructure that can handle the kind of query load that AI applications generate. Jeremy came in with specific knowledge of how to identify quality sources and surface relevant content - the exact problem that plagued every attempt to build AI writing assistants or research tools before retrieval infrastructure caught up with model capability.

By May 2026, Exa had raised $250 million in a Series C led by Andreessen Horowitz, valuing the company at $2.2 billion. That's more than triple the $700 million valuation from the Series B in September 2025. Jeremy was part of the team through both milestones, working on retrieval systems at a company whose growth rate was becoming difficult to explain with ordinary language.

Browser Buddy: A Problem Worth Solving

The premise behind Browser Buddy was a genuine complaint dressed up as a product: the internet's best writing - the patient essays, the technical deep dives, the personal blogs written by people who know things - was getting buried under content optimized for clicks. A recommendation engine that curated for quality rather than engagement was both a tool people wanted and a harder technical problem than it first appeared.

Building it required thinking carefully about what "quality" meant in a crawlable, indexable way. What signals distinguish a thoughtful blog post from a viral opinion piece? How do you build a retrieval system that surfaces writing worth reading, not just writing that was recently popular? These questions sit at the intersection of information theory, machine learning, and content understanding - the exact space Jeremy had been thinking about since university.

His academic background at the University of Virginia included serious engagement with machine learning and information theory. When he built Browser Buddy, he wasn't working adjacent to his interests - he was implementing them directly. And when Exa acquired the company, they weren't just buying two engineers. They were buying a demonstrated approach to a retrieval problem that every AI company needed to solve.

The Writing That Runs Alongside the Engineering

Jeremy maintains a Substack - tagline: "Creating for the long tail." The essays range across technology, philosophy, and personal development in the way that suggests someone who reads widely and thinks in public. The titles give a sense of the range: "Letter to Shareholders (8/26/25)" has the dry corporate form turned sideways; "Irony Makes for Greatness" from February 2026 takes on something more philosophical.

That tagline - "creating for the long tail" - is doing double duty. It's a statement about audience (he writes for the people who will actually care about it, not the masses), and it's a statement about the kind of content worth building for. It's the same instinct that produced Browser Buddy: a conviction that the most interesting things happen at the edges of popularity curves, not at the center.

He also served as a judge at PatriotHacks 2025, a hackathon where YC-backed founders evaluated projects in the AI and Startup categories. The move from participant to evaluator follows a natural arc - someone who built through YC and got acquired bringing that experience back to the next cohort of builders.

The Interests That Feed the Work

Jeremy reads science fiction. He lifts weights. He plays poker. He recently took up latte art and started robotics projects. The range is not incidental. Poker rewards pattern recognition, probabilistic thinking, and patience with uncertainty - skills that transfer cleanly to building systems that retrieve and rank information under ambiguous conditions. Robotics engages the physical side of the systems-thinking that machine learning makes abstract. Science fiction is, among other things, a genre about what information retrieval systems eventually become.

The /weights section on his personal website (jsuh.me) is a pun that holds two meanings simultaneously: the lifting hobby and the model parameters that define what neural networks know. That kind of double meaning isn't accidental. Jeremy Suh is someone who thinks about information at multiple levels - physical, symbolic, and computational - and the personal site reflects it.

What Comes Next

Exa is scaling toward something that hasn't existed before: a search layer purpose-built for AI applications, operating at the scale of the entire web, fast enough for real-time AI inference. The company Jeremy is part of serves enterprises, research institutions, developers, and AI companies who need their models to know what's actually true right now - not what was in the training data eighteen months ago.

His aspirations, as publicly visible, are direct: build new types of retrieval systems that can organize and surface all of the world's information. Browser Buddy was one proof of concept. Exa at $2.2 billion is the larger version of the same experiment. What Jeremy discovered building an essay recommendation engine - that retrieval quality is the binding constraint on what AI can actually do - turned out to be the central insight of a company worth over two billion dollars.

He is 2026's version of the right person at the right place at the right time. Except he helped make the place what it is.