Safi Shamsi

Founder and CEO, Graphify Labs

Please tell us about your current role and employer.

I have believed for a long time that memory is the missing layer of machine intelligence. An AI that reads your codebase and forgets it an hour later is not intelligent in any sense that matters to an engineer. It is a very fast stranger. That conviction began as my MSc thesis at Birmingham, on knowledge-graph retrieval. In April 2026 it became Graphify: an open-source tool, built in a 48-hour sprint, that turns any codebase into a knowledge graph AI coding assistants can query instead of re-reading. The conviction turned out to be shared. Graphify crossed 100,000 GitHub stars in its first 121 days, has passed 4 million downloads, and is now one of the five most-starred open-source projects in Y Combinator's history, maintained by more than 180 contributors around the world. I set the thinking behind it down in a book, The Memory Layer, on building persistent memory for AI agents with knowledge graphs. Today I am the founder and CEO of Graphify Labs (Y Combinator S26). We shipped our first commercial product this month, and Graphify Enterprise, built around formal verification, is onboarding its first design partners. The whole company is, in a sense, the final chapter of my thesis, still being written.

What does a typical day in your current role look like?

Most days mix open-source maintenance, reviewing pull requests from a contributor community that has grown past 180 people, shipping releases (more than 150 since April, at times more than one a day), and responding to developer feedback, with enterprise work that includes conversations with engineers at Fortune 500 companies on our waitlist. I also spend time on research and writing, and right now, preparing for Y Combinator's Demo Day. No two days are identical, but the throughline is always building

What was your journey to getting a job after graduating from Birmingham?

Unconventional, and longer than it looks. I wrote my first line of code at eleven, in a small town near Delhi. After Birmingham I joined Valent Projects as an AI Engineer, working on the Ariadne disinformation-detection platform, tracing how coordinated networks push a false story until it looks organic. Watching machines be confidently wrong, complete with sources that did not exist, shaped everything I have built since. Then, over the Easter break of 2026, I was at home writing The Memory Layer when Andrej Karpathy posted about AI needing persistent knowledge bases. The architecture was already in my head, because it was my thesis. I shipped the first version of Graphify that weekend. When it went viral, I went full-time on Graphify Labs, and we were accepted into Y Combinator shortly after, becoming their first company to cross 100,000 GitHub stars during the programme. The visa process as an international founder was complex and needed careful planning, but the same is true of most things worth doing

How do you feel the learning experience on your course helped to prepare you for your career?

Directly and specifically. My MSc thesis on knowledge-graph-based hybrid retrieval systems was not an academic exercise: it is the technical foundation of everything I have built since, and the formal verification work at the heart of Graphify Enterprise descends directly from that research training. The supervision I received at Birmingham from Professor Paolo Missier and Dr Anelia Kurteva gave me the rigour to publish at MICAD 2025 in Springer Lecture Notes in Electrical Engineering, and to build systems now used by developers at some of the world's largest companies. I am genuinely grateful for that grounding.

How did the University's support services help to prepare you for your career?

The academic supervision was the most formative support. The research environment pushed me to engage seriously with the literature, think clearly about novel contributions, and communicate complex ideas precisely: skills that translate directly into technical writing, open-source leadership, and investor communication.

What are your top tips for students graduating today?

Build something real as early as possible, even if it is small: the gap between knowing a technology and deploying it in the world is where the learning actually happens. Publish your work, whether papers, open-source projects, or writing, because a public record of your thinking compounds in ways you cannot predict. And do not underestimate what a Birmingham MSc is worth internationally. The technical depth of the programme is genuinely competitive at a global level: I have tested that claim in the hardest market there is.

What were the best things about your course?

The thesis. Having the space and supervision to pursue a genuinely novel research question over an extended period, and to own the output completely, was unlike anything else in my education. The programme gave me both the freedom and the structure to do serious work, and I use what it taught me every single day.

What was your biggest achievement during your course?

My oral presentation at MICAD 2025, published in Springer Lecture Notes in Electrical Engineering, on multimodal knowledge-graph-augmented retrieval for diagnosing mycetoma, a neglected tropical disease that must be identified as fungal or bacterial before anyone can treat it: a call that is slow to make and, made late, can cost a patient a limb. The knowledge graph held the record of past cases, so a clinician could see what similar patients had turned out to have rather than trusting a black box. It was the first time I presented original research to an international academic audience, and it confirmed that the work I was doing at Birmingham had genuine scientific merit. Everything since has been built on that confidence.