Inside Katalyze AI: How Reza Farahani Is Rebuilding the Path From Molecule to Market

Katalyze AI is built around a narrow but expensive problem: the gap between a molecule leaving drug discovery and a drug reaching a patient. While most of the industry's attention goes to AI-driven discovery, Katalyze has staked its business on everything that happens afterward, including the manufacturing, supply chain, R&D, and clinical trial work that determines whether a discovered molecule ever reaches the market, and at what cost. Founder and CEO Reza Farahani discussed how he built the company, what its recent $10.5 million seed round unlocks, and why he believes the AI-native advantage in pharma will come not from replacing scientists, but from replicating the best of them.

Who Farahani Is

Reza Farahani is the founder and CEO of Katalyze, a company building agentic infrastructure for life sciences: artificial intelligence applied to the manufacturing and supply chain side of pharmaceutical production. He began his career as a data scientist, moved into consulting at BCG, and then founded and sold his first startup, WF Homey. He launched Katalyze two and a half years ago and has since built a team split between San Francisco and Toronto. The company recently closed a $10.5 million seed round led by Bonfire Ventures.

The Problem: What Happens After Discovery

Farahani starts from an observation he believes the industry consistently overlooks: most of the value large pharma companies create doesn't come from discovering molecules. Of the top 20 pharmaceutical companies, he estimates six or seven haven't brought a molecule of their own discovery to market in the last 20 years.

"They bought companies that took it to market, but they didn't have a molecule discovered."

What big pharma is actually good at is carrying a molecule from discovery to market: manufacturing, scaling, and navigating regulatory approval. That, not discovery, is the core competency. That process is expensive by design. Heavy regulation demands specialized, costly talent, including PhDs, scientists, and engineers, along with facilities many biotechs simply can't afford, which is why so many smaller biotechs are ultimately built to be acquired. Farahani is quick to defend the regulation itself.

"There's something that we are putting in our bodies and we are hoping that cures us. We want to make sure nothing happens, and that it definitely does the job."

Katalyze's goal is to bring down both the cost and the timeline of that post-discovery process, using AI to replicate the science, engineering, and data analysis work that scientists and engineers currently do by hand. Farahani sees the need becoming more urgent, not less, as AI accelerates discovery itself.

"Even Anthropic, Google, everyone is going after that market," he said. "We are going to see a surge of molecules coming to market."

The infrastructure to carry those molecules from discovery to patients has been comparatively underinvested, and that is the gap Katalyze is built to close.

Building the Team, and What Corporate Life Taught Him About Enterprise Sales

Farahani founded Katalyze two and a half years ago with Head of Engineering Matt Cruz, whose experience integrating low-level manufacturing data at Siemens anchors the company's technical foundation. From there, he recruited for a rare and specific profile: people with real experience at the intersection of AI and pharma. The search led to Hannes, a co-founder of Deep Genomics who had built AI companies and worked in research labs at the University of Toronto, and to Shreyas Becker, who joined as COO from Sanofi, where he had sat on the customer side and been responsible for scaling the very product Katalyze was building. Together, the founding team pairs people who understand the industry problem, the AI solution, and the low-level systems integration required to actually ship it, a combination Farahani considers a precondition for the company's success.

That founding mix reflects lessons from Farahani's own path through data science, BCG, and a previous startup. He pushes back on the founder instinct to treat slow-moving, heavily regulated systems as simply broken.

"Sometimes when you come to a regulated environment, that's intentional," he said, "because you are prioritizing not making a mistake over innovation, for a good reason."

Banking is his go-to example: customers don't want the flashiest interface, they want every transaction to work. That understanding changes how you build, engineering safeguards in from the start rather than moving fast and breaking things, an approach he considers right for an innovation lab but wrong for enterprise software. Enterprise buyers, he adds, are fundamentally return-driven. "It's not about the shiny thing," he said. "It's about how this product brings me value."

For Farahani, every enterprise sale reduces to one question: does the product increase revenue or reduce cost? "There is nothing else to it."

The Seed Round, and Why Bonfire Ventures

Katalyze bootstrapped before raising, and the decision to take outside capital was driven by customer demand outpacing what the company could deliver. Of the roughly 200 to 500 discrete operations and workflows involved in taking a drug to market, Katalyze was covering about 20. The $10.5 million seed round funds that expansion: deeper coverage within manufacturing and supply chain, and broader reach into R&D and clinical trials.

The capital has two main destinations. The first is hiring industry veterans, PhDs with 20 to 30 years of experience in a specific role, to consult directly on the product and embed their expertise into it. The second is expanding the AI engineering team, with particular focus on what Farahani calls "harness engineering" and the "context layer": the systems that keep a company's operational data current in near real time, and the infrastructure required to orchestrate AI agents safely, including read/write access and permissions.

"That itself is going to be another topic that we need to focus on more in the next eight to 12 months," he said.

Bonfire Ventures led the round, and Farahani credits the choice partly to partner Brett Queener, a B2B SaaS veteran from Salesforce with direct experience helping startups navigate the "asymmetric interaction" of selling into massive enterprise organizations. He also points to the firm's early conviction on an idea central to Katalyze's own architecture. "They were one of the first people that talked about the importance of the context layer," he said, "and how that's going to create the next era of startups to become successful in the market."

In his view, that positions Bonfire as a genuine thought leader in the space, not simply a source of capital. Selling into pharma also means clearing compliance bars before a single pitch.

"You start with a proof of concept, you show that things work, you show that things have a level of quality." 

That's alongside the certifications and processes regulated buyers expect. He notes that AI is making some of that infrastructure, like privacy and security tooling, less expensive for startups to stand up than it used to be. Katalyze's first major relationship, with Sanofi, became the reference case that has opened doors with subsequent customers.

"You build that trust over time, working with different organizations."

That trust-building plays out concretely in how a deployment starts. New customers connect their data sources, and within the first week or two Katalyze begins building their context layer. "Think about it like it's a data lake for agents," Farahani said.

From there, the company onboards a group of users around specific early use cases, such as yield optimization or a targeted investigation, and works them into using Katalyze in their daily workflow. Usage tends to grow organically from there, though some hands-on onboarding is still typical, less because the product is hard to learn, Farahani says, than because the behavior change it asks of users is significant.

Product, Differentiation, and Competition

Big pharma companies have substantial internal data science teams and no shortage of AI vendors pitching them. Farahani's explanation for why Katalyze wins those conversations is blunt. "The non-technical answer is just it works."

Roughly four months before this interview, a prospective customer told the company that Katalyze was the first AI product they had seen actually function as promised, after a long string of demos that didn't hold up. The industry's core problem, as Farahani sees it, is that AI demos are easy to build and extremely difficult to scale, a gap Katalyze has closed by going deep on a single vertical rather than building something horizontal.

"If you plug our product into a legal solution, it's not going to do the job," he said.

He also credits the team's domain depth, leaders who previously built AI products inside Sanofi and Deep Genomics, with shaping what the product actually needed to do well. Asked for the single piece of feedback that most changed Katalyze's trajectory, Farahani points less to a specific piece of advice than to a disciplined focus on core function over polish.

"It's not about pixel-perfect," he said. "Do we have the most feature-complete product in the market? No. But do we have something that the user sees and says, 'wow, this is great'? Yes."

A user can tolerate a misplaced logout button; they cannot tolerate a result that doesn't hold up scientifically. Concentrating engineering effort on that core scientific accuracy, rather than surface-level polish, was the change that mattered most. Farahani doesn't see other AI vendors as Katalyze's primary competition. The more persistent threat comes from consulting firms pushing pharma companies to build similar tools internally, an arrangement whose incentives, he suggests, can reward slow, open-ended engagements over shipped solutions.

"What's the best-case scenario? The build never finishes, and they keep building it," he said.

The deeper competitor, he argues, is skepticism itself: teams that attempt an internal build, don't finish it, and conclude that the AI doesn't work rather than that the effort was mismanaged.

"It does work, if you choose the right partner for it."

Technical Challenges and the Team Behind Them

Day to day, the company's hardest ongoing challenge is harness engineering and keeping the context layer current. Farahani points to a structural tendency of large language models to gravitate toward the average of their training data, a real liability in pharma manufacturing, where public information is sparse and the goal is to replicate the judgment of the field's most skilled practitioners, not its median.

"I want to replicate the top 2% of scientists," he said, "rather than the median of the population."

Because the underlying economics are so large, even a 2% accuracy improvement on a $10 million production line can translate into roughly $200 million in savings, cost is not a top priority for Katalyze relative to accuracy. Given how quickly the underlying AI skill set is shifting, Katalyze hires for first-principles thinking over surface familiarity with the latest tools: a solid grounding in the fundamentals of AI, statistics, and programming, with enough depth to adapt when the underlying models or techniques change.

"It's so easy to make demos," he said, "but to scale, you need a depth of knowledge."

The hiring bar is whether someone understands the fundamentals well enough to modify a solution as the company scales, not whether they can produce something that looks impressive in a demo. He applies the same standard to his own curiosity about emerging tools and ideas, staying informed about them while remaining anchored to a clear problem so that curiosity doesn't become distraction. For Katalyze, that anchor is concrete: a scientist shouldn't need three months to analyze and report on something as small as changing a light bulb in a facility, when the answer should take a matter of hours.

Scaling the Company, and Himself

Farahani's own day-to-day involvement has shifted meaningfully since the raise. Early on, he was hands-on across sales calls, daily standups, and essentially every function in the company. Now he works deliberately to avoid becoming a bottleneck, while still going deep on a specific problem before pulling back out.

"It doesn't mean you are watching everything from a bird's-eye view," he said. "Sometimes you need to go deep and come back up and move on."

That cadence, he says, keeps his understanding of the problems his team is solving current, without requiring his sign-off on everything. The team itself is split by design between Toronto and San Francisco. Engineering and part of the AI team are based in Toronto, where the University of Toronto and the University of Waterloo, plus researchers like Geoffrey Hinton, give the company access to strong technical talent without competing head-on in San Francisco's hiring market. Product, marketing, and the rest of the AI team are based in San Francisco, where proximity to peers at companies like Anthropic and OpenAI provides a real-time, informal read on where the technology is headed next.

"A lot of learning happens by osmosis," he said.

The same instinct, rebuild rather than patch, shapes how Farahani thinks about the rest of the industry's infrastructure. Asked what belief in enterprise AI or pharma tech he disagrees with, he points to the assumption that legacy systems can simply be adapted to AI rather than rebuilt. Katalyze rewrote its own codebase roughly a year into the company's life, moving from thinking in terms of a data lake to thinking in terms of a context layer, and he expects the same reckoning to hit much larger, older vendors that built their businesses inside pharma's regulatory comfort zone.

"Either these companies substantially change their product, which is very unlikely, or they're going to die," he said.

He expects a broader consolidation of enterprise software as a result, arguing that the regulatory moat many legacy vendors have relied on is not going to hold up in the long term.

Defining Success, and Advice for Founders

Katalyze's stated goal is to cut both the cost and the timeline of bringing a molecule to market in half. That process currently takes roughly a decade and can cost several billion dollars, depending on whether the molecule is small or large, and Farahani estimates that around 60% of that cost and time comes from operational analysis, which he considers a highly targetable problem. Early results are encouraging: work that used to take months or years is now being completed in weeks.

"If that problem gets solved, I don't need to be worried much about myself and the company, how big is it, or stuff like that. If we solve that problem, we are all set."

Asked what advice he'd give a technical founder selling into a highly regulated industry for the first time, Farahani's answer is that trust is never built in a single move.

"These are not, if anyone talks about game theory, a one-shot move," he said.

Start with smaller commitments, make sure each one succeeds, and use that track record to earn room for bolder moves later. Building that credibility with pharma executives took him roughly a year of direct outreach after selling his previous company, drawing in part on relationships from his time at BCG.

"You have that conversation once, twice, five times in a year," he said. "You start building the trust; there is substance to this one."