Every research student knows the feeling. You are early in a project, trying to lock down a hypothesis in your field. You spend weeks scrolling through Google Scholar — and in the worst case, you even start running trial experiments — only to stumble on a paper that already tested and published your idea ten or fifteen years ago, sometimes in another language. By then you have burned weeks chasing citations across a hundred tabs in Connected Papers, scanned recommendations in Research Rabbit, dug through bioRxiv preprints, and checked conference abstracts and threads on Bluesky.
This isn’t a problem of research productivity. It’s a massive infrastructure gap in science.
The Science Infrastructure Problem Nobody Talks About
There are over 30 million researchers worldwide, and more than 100,000 scientific conferences happen every year, spanning 150+ disciplines and 50,000+ institutions — publishing in different languages, formats, and incompatible databases, with no shared platform for live research updates.
We pour our time into keyword searches that are rarely foolproof. Imagine instead searching the world’s collective knowledge through a precise, richly mapped graph — a living, real-time graph that shows who is working on what, with whom, and how ideas are evolving. Today that strategy lives only in the intuition of senior researchers, built over decades of reading, and a new PhD has no way to inherit it. Axy is built to close this gap — the version-controlled, shared layer that science is still missing.
Software engineering was chaotic before GitHub, too: code lived on local drives, and collaboration meant emailing ZIP files and guessing. GitHub fixed that with versioned repositories, pull requests, and a shared layer — and open source exploded. Here is how Axy brings that same shift to research.
What Is a Knowledge Graph?
Without getting lost in jargon, think of it as a simple graph made of nodes, which are entities (things), and edges, which are relationships (connections between things).
A scientific knowledge graph extends this idea into research. The nodes are conceptual entities (hippocampus, SARS-CoV-2), processes (translation, synaptic plasticity), properties (binding affinity, immunogenicity), methods (optogenetics, CRISPR, fMRI), models (Hodgkin–Huxley, the SIR model), or phenomena (long COVID, the placebo effect). The edges are the relationships between those concepts, and each edge is backed by the papers that establish it — evidence showing whether one result supports, extends, or contradicts another.
Connect the nodes through their relationships and you get a mapped network of papers and citations that is genuinely alive and answers your query effortlessly. This is a world apart from a keyword search for — “Papers on memory consolidation” — which can now be refined to ask: “Studies connecting hippocampal replay, systems consolidation, and sleep-dependent memory reactivation that use in vivo calcium imaging but report conflicting interpretations.” A search like that hands you exactly the papers you should read, including the ones a PubMed search would never think to link together.
This also beats plain citations. When Paper A cites Paper B, all you learn is that they are related — not whether A supports B, extends B, or contradicts it. Consider the work of Katalin Karikó: her early mRNA research took years to influence later breakthroughs, largely because direct citation paths missed the concept-level relationships. A knowledge graph would surface papers that never cite one another but share the same question, methods, or assumptions — the kind you might extend to a new model species. Knowledge graphs use machine learning to infer edges from text, and then humans — via apps like Axy — refine them.
Axy’s pilot with Prof. Gruber at the University of Geneva shows the difference. His team studies primate cognition, so their graph has nodes for primate species, behavioural traits (tool use, social learning, mirror self-recognition), and emotional states (fear, curiosity, joy). With continuous use, Axy’s AI sharpened the edges between those nodes — which species show which behaviours, under what conditions, with what confidence, and from which studies. A literature survey that would take weeks of reading and hard-won intuition now takes seconds, and it surfaces papers close to your question that otherwise wouldn’t appear for months.
”GitHub for Science”: Why the Analogy Matters
In the early 2000s, software had an infrastructure problem that feels remarkably like science’s today. GitHub didn’t invent a new way to code; it changed how code was stored, connected, and shared. With version control, you could fork a project, track every change, see exactly who contributed what and when, compare versions, and merge insights from different contributors. You could stand — visibly and traceably — on the shoulders of everyone who came before. Isn’t that the real essence of research?
Because science still lacks this connectivity, it is in many ways stuck in the pre-GitHub era. Hypotheses evolve. Methods improve. A research question posed in a 2008 paper might have seeded six labs across three continents over the following decade, each interpreting it differently. Because none of them are connected, that evolution exists only as a trail of fragmented papers, tacit lab knowledge, and disconnected discussions. There is no fork, no merge, and zero version history.
Version control for scientific ideas would look like a concrete, branching map: if a hypothesis grows out of an earlier theoretical framework, you could track how it was refined, what challenged it, who refined it, which experiments supported it, and the intellectual lineage of an idea across institutions and decades. And every version becomes comparable. When two labs reach different conclusions about the same question, the divergence lands on the map. When independent lines of research converge on the same answer, the graph makes that convergence visible.
This is a literal rendering of Newton’s line, “If I have seen further, it is by standing on the shoulders of giants.” As the graph is fed continuously, PhDs can onboard instantly and PIs can spot collaboration forks. Axy’s app is the seed for this: conference talks that tag concepts and journal-club votes continually refine the graph.
How Knowledge Graphs Could Revolutionize Research
This is where the idea turns real — when you unlock a structured, connected, queryable map of scientific knowledge.
1. Hidden connections surface
Often a paper is indirectly relevant and could help enormously, but you never know it exists — and if it is lightly cited, you would never go looking for it. Return to Karikó: dozens of papers built on her 2005 mRNA work (the research behind her 2023 Nobel Prize), yet none cited it, because the connection lived only at the level of ideas. A graph spots that connection and bridges those papers by asking what each is about, what questions it answers, and where else those questions are being asked.
2. Research gaps become visible
It takes years to develop the instinct for spotting a research gap; a senior researcher’s intuition works because they have read a field from many angles. For an early-career researcher, knowledge graphs change everything. Google Scholar and a knowledge graph can both hand you 12 papers on X (say, tool use) and 8 on Y (say, emotions) — but only the graph shows you the white space between them, the missing link through Z (perhaps social learning). That visible gap is where your next paper lives.
3. Literature reviews go from weeks to hours
A thorough literature review traditionally means weeks of reading hundreds of papers. With a knowledge graph you still read — but the graph transforms which papers you choose first, building the bridges that reveal the gap.
4. Smarter AI for science
Plenty of people are dazzled by large language models; others find them useless past a point. The reason is simple: they hallucinate, confidently inventing claims by matching patterns in text without understanding it. Knowledge graphs are different. Built on structured knowledge and paired with generative AI, they don’t just sound scientific — they can reason over real claims, relationships, and chains of evidence.
5. Institutional memory for early-career researchers
When a senior PhD or researcher leaves, their knowledge leaves with them — which paper is foundational versus incremental, who is working on what, what was tried before and why it failed. A knowledge graph changes this: that knowledge is saved as institutional memory, fed continuously, and available for everyone who comes next.
6. Real-time collaboration gets smarter
You meet more people and encounter more relevant work when the internet connects you by shared ideas rather than by broad field labels. Knowledge graphs connect you with researchers who share concepts even if you have never been in the same room or at the same conference.
Why This Hasn’t Been Built Before — And Why It Can Be Now
The tools we already use do their jobs well, but they hit the same wall: you cannot manually and accurately tag a million papers. Google Scholar and Semantic Scholar are excellent search engines, but they can’t capture the conceptual relationship between two papers. Connected Papers and ResearchRabbit build visualisations from papers that cite each other. They can place a paper in a domain; they can’t establish the relationship between them.
Graphs need deep knowledge of which ideas connect to which, and which concepts and methods span across time — and that can only come from experienced researchers. Even a room full of hired domain experts can’t keep up with years of research across a growing set of fields.
Axy is built to capture the working memory of real scientific interactions as they happen. When a scientist uses the Axy app during a conference talk, it generates a signal about which ideas matter and creates a node. When a lab records its journal-club discussion or votes on a paper, it builds edges — learning how one paper connects to another, which claims are challenged, and which methods are adapted. The data isn’t fed at random; it forms a living scientific graph in real time.
This is what turns a user base into a data flywheel. Every application, every tagged talk, every journal-club vote makes the graph richer; a richer graph makes the tools smarter; smarter tools welcome more researchers, who in turn add to the map. Axy is building that flywheel now, cohort by cohort.
That is why executing on knowledge graphs is finally possible — not because of a single technological leap, but because there is finally a platform researchers can use in their daily work and benefit from as they do.
How Axy Is Betting on the Future of Research Infrastructure
If this works, scientific literature search will look completely different within a decade.
A new researcher won’t dig into a field from scratch over four days — they will simply ask the graph, and get back what has already been done, who is working on similar questions, where the gaps are, what methods have been tried, and what results came out. Imagine the time, focus, and intelligence that could then be redirected toward genuinely new advances.
It would also mean researchers from entirely different fields — say materials science and synthetic biology — could see their shared ground and the gaps still open between them. Someone modelling climate change might draw an idea from a fluid-dynamics paper by way of atmospheric chemistry. Two labs on different continents working on a similar mechanism would share nodes on the same graph, making collaboration by idea-pooling far likelier than by chance collision.
Alone, neither AI nor a human can do these thousands of hours of work. Together, the power compounds. AI stops hallucinating because it draws from a knowledge graph and reads real claims, relationships, and chains of evidence.
The next leap in science may not come from a new instrument or sequencing technology, but from the infrastructure that lets researchers share and build on each other’s work. Right now, most of us connect only within our existing network, drawing on the handful of papers and talks we have personally encountered. A scientific knowledge graph changes that: every shoulder gets its credit and becomes visible, every connection is mapped, and finding a research gap becomes a smooth, everyday process.
Axy is building this graph now — starting with a conference app and journal clubs as its first instruments — and we’re inviting our first 500 researchers in small cohorts. If you want to help build the shared map of science rather than just use another tool, apply to join below.
