Perceptic

Tilman Flock is one of a small number of people who have published landmark research in structural biology and then gone on to build production enterprise AI at scale. He is a computational and molecular biologist who completed his research fellowship at Fitzwilliam College, University of Cambridge, supervised by Madan Babu at the MRC Laboratory of Molecular Biology. His doctoral and postdoctoral work focused on G protein-coupled receptors (GPCRs), the single largest class of drug targets in medicine: roughly a third of all approved drugs work by binding to GPCRs. His papers, published in Nature and co-authored with Frances Arnold (2018 Nobel Laureate in Chemistry) and other leaders in structural biology, have been cited over 2,800 times. The most-cited work described a universal allosteric mechanism for how GPCRs activate G proteins, a foundational result for anyone designing drugs that target this receptor class. Flock's intellectual arc is the direct predecessor to Perceptic. He spent years at Cambridge asking: how do GPCRs selectively engage the right signaling partners at the molecular level? At Palantir, the same question took a different form: how do fragmented enterprise data systems selectively connect insights to the right decision-makers? The computational biology trained him to think about selectivity, specificity, and the cost of missed connections in complex systems. Palantir gave him the production AI infrastructure to solve those problems at enterprise scale. He spent nearly seven years at Palantir, building AIP and running the Life Sciences practice. He was the person who sat in rooms with pharma executives and translated what LLMs could actually do in production inside regulated data environments. His Fortune interview is the most complete public record of his product thesis: "Too many critical drug development decisions still happen without a complete view of the evidence." That is not a market observation. It is the frustration of someone who has watched it happen up close for seven years. His Google Scholar profile lists 24 published papers with sustained citation growth through 2025 and 2026, despite having been in industry since 2019. That means his academic work is still generating downstream research from the scientific community, even though he has moved on. He is listed as a former Palantir employee with a verified institutional email from his Cambridge fellowship days, a small detail that reflects a researcher who kept both identities active even as he shifted domains. He is based in Switzerland, which gives Perceptic's Basel presence its anchor. What he talks about publicly: Flock is not a frequent social media poster. His public intellectual output is concentrated in the papers and in the handful of founder interviews he has given since stealth exit. His Fortune interview reveals a specific frustration: the AI drug discovery community has been building better instruments at individual stages of the pipeline, while the decision-making architecture connecting those stages remains broken. He does not name competitors carelessly. He named the problem. Lesser-known facts: His published research includes a collaboration with Frances Arnold's lab at Caltech on cofactor specificity switching in oxidoreductases, an enzyme engineering problem with direct implications for industrial biotechnology. Arnold won the 2018 Nobel for directed evolution. Flock's work with her lab is a quiet signal about the range of his scientific collaborations and the level at which he was operating as a computational biologist. His Fitzwilliam College fellowship was at one of Cambridge's oldest colleges, the same institution that produced the architects of the Human Genome Project. He left Cambridge research for Palantir not because the science was done, but because he could see that the production AI infrastructure problem was going to determine whether the science could ever be used properly inside pharma.

Martin Copes started coding at age 12 in Uruguay, sneaking out of bed at night to build an online tennis game on a heavy CRT monitor. He released it to tens of thousands of players worldwide before he was a teenager. That is the founding data point: he is a builder for whom the feedback loop of releasing something into the world is a genuine physiological need, not a career strategy. He describes the feeling of watching players react to his game as "the feeling I've been chasing since." The second time he felt it was getting his hands on GPT-2. He is a biomedical researcher who made a sharp pivot into production AI. His academic background is in biomedical research, specifically studying how AI can be applied to understand complex biological systems. He was, in his own words, grinding all-nighters over biological data before he discovered large language models. When he encountered GPT-2, he treated it the way he had treated his childhood game engine: as something to be understood from the inside, immediately. That obsession took him to Palantir in 2022. At Palantir, he and Zaki Trache were pulled in personally by Shyam Sankar, Palantir's CTO, to build the technical foundations of AIP from its earliest days. They scaled the core LLM services underlying AIP to process millions of reasoning requests per second with fine-grained access controls, work that Copes describes as "long before 'agents' became fashionable." The AIP architecture that every Palantir life sciences customer uses today was built by this founding team. That is not a resume line. It is product proof. His LinkedIn announcement post for Perceptic is the most revealing public document about how he thinks. He describes the biological motivation in precise terms: "Drug discovery isn't just expensive and slow because the science is hard. Even the best researchers can only go so deep in so many directions at once. Critical context gets lost between teams. Institutional knowledge disappears between projects. Promising paths go unexplored. Not for lack of ideas, but for lack of capacity." That is a product thesis written by someone who understands both the science and the software failure modes simultaneously. What he talks about publicly: Copes posts rarely and specifically. His Perceptic announcement post is long, personal, and technically precise. He is not interested in thought leadership. He is interested in showing his work. The combination of biomedical research background and AIP-level production engineering is rare, and he knows it, but he does not advertise it generically. He describes specific decisions, specific scale numbers, specific feelings about what building means. Lesser-known facts: Copes grew up in Uruguay, which puts him in the company of a small but serious cohort of South American technical founders who came up through entirely non-standard pathways into global enterprise AI. He self-taught programming from a CRT monitor in a country where the tech ecosystem was negligible. The game he released to tens of thousands of players worldwide was not a school project. It was an independent product, built by a child, distributed globally. That is the throughline of the founding team: all three of them built things in difficult environments before anyone gave them permission or resources to do it.

Zaki Trache is the third co-founder and, alongside Martin Copes, the person who built the technical foundations of Palantir's AIP. He and Copes were both recruited personally by Shyam Sankar to establish the LLM infrastructure layer that became AIP's core. Copes' LinkedIn post names him specifically: "Zaki and I were pulled in by Shyam Sankar to wire the foundation for Palantir's AIP." That verb, "wire," is precise. They were not the product managers or the strategy team. They were the people who built the circuits that everything else ran on. Perceptic is based in London and Basel. The Basel presence is where Tilman's scientific network and Swiss pharma ecosystem connections live. Zaki and Martin are the London engineering anchor. Together, all three founders bring a combination that is genuinely uncommon: one founder with published Nature-level structural biology research and six years of Palantir life sciences deployment experience, and two founders who built the LLM infrastructure layer that Palantir now sells to the world. Perceptic did not need to go find domain expertise or production AI experience. It was founded by all three simultaneously. Zaki maintains a low public profile outside LinkedIn. He is not a frequent poster or public intellectual in the way Flock has become in the life sciences AI community. His work is in the architecture. The best public evidence of his thinking is the product itself, which has achieved a 50-fold increase in clinical data extractions in production and is scaling asset scouting from hundreds per week to thousands in minutes. What he talks about publicly: Minimal public presence. LinkedIn is the primary surface. His contributions to Perceptic's technical direction are documented through the product outcomes and through Copes' and Flock's public descriptions of the AIP-building work. Lesser-known facts: Being personally recruited by Shyam Sankar is not a standard hiring event at Palantir. Sankar is Palantir's CTO and the person most directly responsible for AIP's technical vision. He does not pull in engineers personally for generic hires. The fact that both Zaki and Martin were recruited this way, before AIP was what it is now, and that they stayed to build the core LLM infrastructure layer from scratch, means they have worked on the hardest part of what is now Palantir's most commercially important product. That is the engineering pedigree Perceptic is building on.