Latent Knowledge

There is a specific kind of despair that belongs to the research literature review, and anyone who has done one knows it. You type words into a box. The box returns ten thousand documents ranked by an algorithm that was optimized for something - relevance, citations, recency - but almost certainly not for the thing you actually need, which is to understand a field you don't yet understand. So you read. You open tabs. You lose the thread. And the paper that would have changed your mind is sitting on page 47, which you will never reach, because no one reaches page 47.

Latent Knowledge, a company founded in 2019 in New York by James Reilly, is a bet that this is a solvable problem, and that the solution is not a better ranking of the list but the abolition of the list. Their product, LitView, does something that sounds modest and turns out to be a small philosophical position: instead of returning documents in a line, it returns them in a map. It clusters research literature by meaning, groups related work together, and lets you see the shape of a field at a glance - the neighborhoods, the gaps, the papers sitting quietly between two topics that no keyword would ever have connected.

The technical term for the machinery underneath is applied natural language processing, and the marketing term is "AI-powered," which in 2019 was a bolder thing to say than it is now and in 2026 is a thing everyone says. What distinguishes Latent Knowledge is not that it uses NLP - so does your email spam filter - but where it points it. The company aimed a genuinely hard technology at a genuinely narrow, genuinely painful workflow: how research and development teams find the literature they need. Narrowness, in startups, is usually a virtue disguised as a limitation.

The founder was the user

The most reliable origin story in software is "someone needed a tool that didn't exist and got tired of waiting." James Reilly's fits the template unusually well. Alongside building Latent Knowledge, he worked as a researcher in the life-science program at the United States Military Academy at West Point - which is to say he was, personally, one of the people drowning in literature that LitView is designed to rescue. This is not a small biographical detail. Products built by their own users tend to have opinions, and opinions, in product design, are the difference between a tool and a toy.

It also explains the company's instinct for where to plant the flag first. LitView's early deployments were not consumer-friendly, high-volume, low-stakes places. They were high-credibility, high-stakes ones: West Point itself, and researchers connected to the Global Alliance for Preventing Pandemics at Columbia University's Mailman School of Public Health. These are users who cannot afford to miss a paper. A tool that survives contact with people whose work is measured in lives is a tool you can then sell more or less anywhere.

What LitView actually does

Strip away the category language - "literature-based discovery," "semantic search," "research content clustering" - and LitView does three concrete things that a keyword search cannot. First, it lets you search by an idea or by an existing document, not just by words, which is closer to "find me more like this" than to "match this string." Second, it lets you bring your own private data alongside public literature, so a company's internal research sits in the same map as the published field. Third, it visualizes the result as clusters, so the relationships between disparate technical articles become something you can see rather than something you have to reconstruct in your head.

The pitch to an R&D team is therefore not "search, but faster." It is "search, but you stop missing things." Those are different products. The first sells on convenience and competes on price. The second sells on the fear of the unknown unknown - the paper, the prior art, the adjacent result you didn't know to look for - which is a much better thing to be selling, because it is a fear that never fully goes away.

Small on purpose

Latent Knowledge is not a large company, and reported figures put its team somewhere around 29 to 31 people and its total funding modestly - the numbers across public sources range from a few hundred thousand dollars to just under $700,000, with a seed tranche reported through Techstars in early 2023. In an era when "AI startup" is often shorthand for "raised nine figures to build a chatbot," this is almost defiantly unglamorous. It is also, plausibly, the point.

The company completed the Techstars Tech Central Sydney Accelerator - a New York company that flew to Australia to sharpen itself, which tells you something about how founders chase the right room over the nearest one. It went through the National Science Foundation's I-Corps program, the customer-discovery boot camp that exists specifically to keep technical founders from building things nobody wants. By late 2023 it was being described as having de-risked its business and looking toward growth. None of this is the language of a rocket ship. It is the language of a company trying to be correct before it tries to be big, which is the harder and less fashionable order to do things in.

When Microsoft calls

In 2022, Microsoft agreed to co-sell LitView - to distribute a small New York startup's search engine through its own channels. It is worth being precise about what this does and doesn't mean. It does not mean Microsoft bought them, or blessed them, or guaranteed anything. Co-sell partnerships are, at bottom, a shared bet: the larger company decides your product is worth putting in front of its customers, and you decide its reach is worth the revenue share. What makes it notable here is the asymmetry. Distribution partnerships of this kind are not usually about a startup's size. They are about whether a big company thinks you're pointed at where the market is going. Someone at Microsoft looked at semantic literature search and decided it was.

The other partnership worth noting is quieter and, in a way, more interesting. Latent Knowledge agreed to power the literature search inside IndyGeneUS AI's Precision Health Discovery platform - meaning LitView stopped being only a product you buy and became infrastructure inside someone else's product. This is a classic route by which small, focused tools become difficult to remove: not by being the thing on the screen, but by being the thing underneath it.

Design thinking, meant literally

Latent Knowledge talks about design thinking, which is normally the kind of phrase that should make you check your wallet. In most companies it means a poster near the coffee machine and a workshop nobody remembers. Here it appears to mean the more literal, more useful version: watching how clinical, biotechnology, and public-health researchers actually work, then removing steps until the tool matches the workflow rather than the other way around. The company's own materials lean on the standard design-thinking statistics - that design-led companies can outperform the broader market, that design-driven organizations report better returns - which is a slightly circular thing for a small startup to cite, but it also tells you where the team's convictions sit. They believe the interface is the product. For a tool whose entire job is to make a large, tangled field legible at a glance, that is close to correct.

Who it's really up against

Latent Knowledge does not operate in empty space. The incumbents of research discovery are formidable and, mostly, free: Google Scholar, Semantic Scholar, Dimensions, and a growing cohort of AI-native tools - Elicit, Scite, Consensus - all competing to make the literature more tractable. Against free and famous, a small paid startup needs a wedge, and Latent Knowledge's is the combination that the incumbents mostly don't offer together: private-plus-public data in one view, visual clustering rather than lists, and a deliberate focus on R&D and life-science teams rather than the whole world of students and academics. It is not trying to be everyone's search engine. It is trying to be the specific tool for the specific person who cannot afford to miss a paper - which is a smaller market, and a stickier one, and a much easier one to charge for.

The bigger idea

There is a broad, slightly under-noticed trend in AI: the more interesting companies are increasingly the ones that map knowledge rather than merely retrieve it. Retrieval gives you a document. Mapping gives you a landscape. LitView is a small, specific, working instance of the second thing - a claim that the future of search looks less like a ranked list and more like a territory you can walk around in. Whether Latent Knowledge is the company that scales that idea or merely one of the early, correct ones to state it is, at this stage, genuinely unknown. But the idea is right, the founder lived the problem, and the tool exists and is used by people who cannot afford to miss a paper. In startups, that is more than most get to say.