Megha Sinha spent close to two decades billing life sciences companies for regulatory planning work — then built software to put that business out of business. In this episode, Matt talks with the founder and CEO of Kolter AI about why she treats regulatory planning as a computation problem rather than an LLM problem, how the platform's architecture evolved after an early PDF-and-RAG approach didn't hold up, the trust model that decides when an AI agent gets to recommend versus decide, and what's left to compete on once an entire industry can run the same planning engine.
Megha Sinha is the founder and CEO of Kolter AI, an autonomous regulatory planning platform for life sciences, and managing partner and CEO of Kamet Consulting Group, which she has run since 2023. She has spent 17 years inside major life sciences lifecycle-change programs — M&A integrations, divestitures, tech transfers, post-merger rebrandings, and CMC changes across global portfolios — across 125+ cross-functional engagements. Eight of those years were at PwC, where she built and led the firm's Life Sciences Regulatory Consulting practice (Health Industries Strategy and Transformation, Global Regulatory Consulting Solutions Leader), growing it from a single capability into a cross-platform business and supporting engagements including a $63B pharmaceutical acquisition, a $25B pharma integration, and a $2.7B divestiture. At Kamet, she has led regulatory strategy for a $14B global healthcare spin-off — rebranding across 100+ countries, 45,000 SKUs, and 10,000+ label redlines — and a $14.6B consumer health carve-out. That spin-off-scale problem, building a plan by hand across tens of thousands of SKUs, became the seed for Kolter AI. Earlier in her career she held regulatory affairs and quality roles at Stryker and MD Resource Corp. She holds an MS in Biomedical Engineering/Regulatory Affairs from the University of Southern California and a BE in Chemical Engineering from R.V. College of Engineering, Bangalore, and is RAC certified.
From billing hours to giving it away. Megha built the first version of what became Kolter as an internal automation engine for her own consulting practice — the kind of tool that could have easily billed a client for a week of work but instead took a matter of hours. Rather than protect that as a competitive edge, she chose to build a product around it, calling it "cannibalizing my own consulting business."
Why the intelligence layer was the hardest part to build. The early architecture put PDFs behind a retrieval-augmented generation (RAG) layer, but Megha found that LLMs don't read documents consistently or deterministically. The team moved to a structured database schema and roughly 38 small, purpose-built computation engines instead — keeping the deterministic math and rules out of the model's hands entirely so the same input reliably produces the same output.
Modular by design. An early beta produced an entire cross-functional plan — regulatory, labeling, supply chain — in one shot. Client pushback ("I don't want to tell supply chain what to do") led Megha to split Kolter into four or five separate agents, so each function can pull only the modules it wants.
Trust in increments. Megha's current framework: administrative decisions get automated outright, genuinely ambiguous regulatory judgment gets escalated to a human, and once a human has approved the same kind of mitigation enough times, the system starts presenting it as a precedent to accept rather than a question to answer from scratch. She frames this as a recursive self-improvement loop, with human-in-the-loop as today's checkpoint rather than a permanent one.
Strategic advantage versus tribal knowledge. If most of the industry ends up running the same planning engine, what's left to compete on? Megha's answer: the mechanics of building a plan quickly stop being a differentiator once they're commoditized. What stays proprietary is company-specific tribal knowledge and negotiated precedent, which customers can feed back into the platform as custom computation logic without a code change.
The cross-functional bottleneck hiding behind regulatory timelines. Megha points to the EMA's pilot on parallel post-approval-change submissions across reference and rest-of-world markets as an example of a shift the industry isn't operationally ready for: if health authorities start approving submissions in parallel, the burden shifts overnight from regulatory drafting to manufacturing and supply-chain readiness — and most organizations have no current line of sight into those downstream implications.
What's next. Megha describes early interest in extending Kolter into clinical — protocol amendments, IRB changes — and eventually letting agentic systems handle regulatory data entry directly, freeing operations teams for higher-value, patient-facing work.
Megha Sinha, on why she thinks regulatory belongs at the strategy table: "I truly believe that the regulatory folks in our industry are the true strategists... we take a drug and asset and build strategy, right? So we are true problem solvers."
Megha Sinha, on the design choice to keep LLMs out of the decision-making: "LLMs are pretty bad in math... I don't want them to make the decision, right? Because I wanted when I give the same input, I get the same output."
Megha Sinha, on choosing to build a product instead of protecting a consulting revenue stream: "I'm cannibalizing my own consulting business in some respect, but the expertise that I've developed over the years that will stay with me... I want to democratize it."
Megha Sinha, on where she draws the line for AI agents today: "If you are not sure, don't make the decision, right?"
Megha Sinha, on the industry's biggest objection to a shared planning engine: "Why would people want to give me their tribal knowledge? You know, that's their secret sauce, right?"
Matt, on why he named the podcast Operations Utopia: "If the patient is the point, and speed is the point, we should not have any excuse. We are obligated to not waste time if we don't have to."
Regulatory operations leaders, RIM and submission-planning teams, life sciences data and technology leaders evaluating AI for compliance-critical workflows, and founders building vertical AI tools for regulated industries.
Guest & Company
Related