For Daniil Boiko and Andrei Tyrin, the idea for Onepot AI came from the same frustration. “The best ideas in drug discovery were often blocked not by biology, but by synthesis,” Boiko told TechCrunch. Synthesis is the creation of new molecules by using chemical reactions. It’s like a recipe or Lego pieces, where small pieces, ingredients, molecules, come together to form a wider puzzle picture, a food dish, a larger molecule. As one might expect, it’s quite hard to create those small molecules that go on to build bigger ones. For Boiko, a PhD candidate studying machine learning in chemistry at Carnegie Mellon (he received his bachelor’s and master’s in organic chemistry from a university in Russia), that meant realizing that drug hunters — the scientists who oversee drug discovery and development — were skipping promising ideas just because the chemical molecules to create the drugs seemed too hard to make. “The compounds never even got a chance to be tested,” Boiko told TechCrunch. For Tyrin (who received his bachelor’s in computer science at MIT), his time working on drug discovery computational pipelines made him realize how behind the world of drug discovery was. “The models could generate ideas in hours, but it could take months for the lab to catch up,” he told TechCrunch. “We both saw that the world was throwing money into molecular design and almost ignoring the harder problem of actually making the molecules,” Boiko said. But there was a geopolitical angle too, he continued, global supply chains are becoming vulnerable, and the U.S. is entering a trade war and innovative competition, once again, with China. Techcrunch event San Francisco | October 13-15, 2026 “It was clear,” Boiko said. “Small molecule synthesis needed to be rebuilt from the ground up in the United States.” Boiko and Tyrin came together to create Onepot, a company that is home to the small-molecule synthesis lab POT-1. They also built Phil, an AI organic chemist, to help run experimental analysis to increase the process of compound synthesis for their early commercial partners. Those partners are biotech and pharma companies that are currently trying out their technology. On Wednesday, the company came out of stealth with <head>3 million of funding, including pre-seed money, and a seed round led by Fifty Years. “Currently, pharma and biotech companies either build entire teams of chemists in-house or work with contract research organizations overseas,” Tryin said of the process for molecular synthesis. Human chemists can spend months of research to create even a single compound, at a cost of thousands of dollars. It involves a lot of trial and error — studying various compounds, collecting data on biological activity, how the drug moves through the body, toxicology reports, and coming up with what to experiment with next. “The main limiting factor here is not testing these compounds, but making them in the first place,” Tyrin continued. “We aim to compress this down to days.” Tyrin said the product is pretty straightforward. Onepot has a catalogue of molecules it can make. Clients choose which compounds it want and then Onepot’s technology will synthesize the molecules and then ship them to the customer so the customer can use them in their own experiments. (They ship physical products either as dry compounds of solutions in plates or vials). The backend of the product is where Boiko and Tyrin have fun, dissecting the problems of chemical synthesis to find out which combinations of molecules work together. They built a lab where they are letting LLM agents access these so-called molecule recipes for training so the agents can also find out what works and doesn’t in compound building. “When executing experiments in the lab, we capture every single detail that goes into the process,” Tyrin said — that means tracing temperature and, essentially, the ingredients that were added to a mixture to create compounds. “No information is lost, which makes experiences reproducible even if someone decides to run them in ten years from now.” This also means their agents generate hypotheses from real-world experiments rather than literature data, often mined from the internet. Boiko called the fundraising process “hectic,” and said they met their lead investor through an intro. “What was supposed to be a short meeting turned into a multi-hour whiteboard session about industrializing synthesis,” Boiko said. Others in the round include Khosla Ventures, Speedinvest, OpenAI co-founder Wojciech Zaremba and Google’s Chief Scientist Jeff Dean. The fresh capital will be used to build a second lab in San Francisco so the team can take on more customers. It will also expand the team and its compound discovery engine. On the service side, Boiko and Tyrin look at WuXi AppTec and Enamine as competitors.