Frekil Turns Months of Clinical Data Analysis Into Minutes

The Macro: Real-World Evidence Is Pharma’s Bottleneck

Every drug that makes it to market carries years of clinical trial data. But trials are artificial environments. They have strict inclusion criteria, controlled conditions, and sample sizes that rarely reflect how a drug performs across millions of diverse patients in the real world. That’s where real-world evidence comes in. RWE pulls from electronic health records, insurance claims, patient registries, and other sources to show how drugs actually perform outside the lab.

The FDA has been pushing hard for RWE. The 21st Century Cures Act gave it formal standing, and the agency has been steadily expanding the contexts where RWE can support regulatory decisions, from label expansions to post-market surveillance. Pharma companies need it. Payers demand it. The problem is that generating RWE is brutally slow and expensive.

A typical RWE study involves pulling data from multiple sources, cleaning and standardizing it (often to the OMOP Common Data Model), writing analysis protocols that meet FDA requirements, executing statistical analyses in R or Python, and producing publication-ready tables, figures, and listings. This process takes months. It requires specialized biostatisticians who are in short supply and command high salaries. Companies like Iqvia, Optum, and Veeva have large RWE practices, but their services are expensive and slow.

Smaller players have chipped at this. Flatiron Health (now owned by Roche) focuses on oncology RWE. Aetion builds a platform for regulators and manufacturers. Truveta aggregates health system data. But the core workflow, the actual analysis pipeline from raw data to regulatory output, remains largely manual and painfully slow.

The Micro: An AI Biostatistician From Stripe and Sony Alumni

Nikhil Tiwari and Shivesh Gupta founded Frekil after engineering stints at Stripe, Amazon, and Sony Japan. Both are IIT Bombay graduates, and they came through YC’s Spring 2025 batch. The team is two people, based in San Francisco.

The product is an RWE automation platform that takes raw clinical data (EHR records, claims data, registry data) and generates regulatory-grade evidence. The pipeline has several distinct steps, and each one represents work that currently takes a specialized human days or weeks.

First, data integration. Frekil connects to EHR systems, claims databases, and registries, then automatically maps the schemas and standardizes everything to OMOP CDM. If you’ve ever tried to reconcile data formats across different hospital systems, you know this step alone can consume weeks of engineering time.

Second, protocol generation. The platform uses AI to draft FDA-aligned study protocols. This is where domain expertise matters enormously. A protocol that doesn’t meet regulatory standards is worthless, and getting it wrong means starting over.

Third, code execution. Frekil generates R and Python analysis code and runs it in a secure sandbox. The emphasis on transparency here is important. Pharma companies and regulators need to audit the analysis code. Black-box results don’t fly in regulatory submissions.

Fourth, output generation. The platform produces publication-ready tables, figures, and listings (TFLs) automatically. In the traditional workflow, this formatting step is often outsourced and takes weeks of back-and-forth.

The most interesting architectural decision is their claim that “no clinical data ever touches AI.” They maintain separation between the AI models and patient records, which addresses the biggest objection any pharmaceutical company or hospital system would raise about using AI in clinical data analysis.

The use cases they list are comprehensive: HEOR and market access, safety and pharmacovigilance, label expansion, competitive intelligence, trial feasibility, and external control arms. Each of these is a distinct buyer within a pharma organization, which suggests they’re thinking about the go-to-market carefully.

The Verdict

I think Frekil is attacking the right problem at the right time. The demand for RWE is growing faster than the supply of biostatisticians, and the regulatory environment is actively encouraging more real-world data in drug evaluations. That’s a structural tailwind.

The technical approach is sound. Automating the pipeline from raw data to regulatory output, while keeping patient data separated from the AI layer, addresses both the efficiency problem and the trust problem simultaneously. The OMOP CDM standardization piece alone could save customers weeks per study.

The risk is credibility. Pharma companies are conservative buyers, especially for anything that touches regulatory submissions. A two-person startup asking a pharmaceutical company to trust its automated analysis for FDA submissions is a hard sell, regardless of how good the technology is. The Stripe and Amazon backgrounds help with engineering credibility but don’t carry weight in biostatistics or regulatory affairs circles.

They’ll likely need strategic partnerships with established RWE providers or a few visible case studies with mid-tier pharma companies before the enterprise buyers take a meeting. The demo-first approach (the site directs to “Request a Demo” with no public pricing) is the right call for this market. Nobody in pharma buys RWE tools off a pricing page.

If the technology delivers on the speed claims and the regulatory outputs pass scrutiny, Frekil could compress a process that currently costs hundreds of thousands of dollars and takes months into something a small team can run in days. That’s the kind of efficiency gain that justifies a premium price and builds deep switching costs. The question is whether they can get enough early customers to prove it before the established players build their own AI layers.