Health tech · Real-world evidence software
Frekil
An AI platform that says it can compress a real-world drug-safety or effectiveness study from a months-long service engagement into a same-day workflow. It was founded in 2025 by two IIT Bombay graduates who started the company doing something else: labeling medical scans.
Chat with your data
The pitch on Frekil’s site is a single line: describe a clinical question in plain language, and the software takes it from there. Connect an electronic health record system, a claims database, or a registry, and a researcher types a hypothesis the way they might type a search query. The platform says it structures the inquiry, builds a patient cohort, drafts a causal diagram, writes a statistical analysis plan, extracts the data, runs the analysis, and writes the report, with a human reviewing each step first.
That eight-stage sequence, literature review, question clarifier, cohort builder, causal DAG builder, SAP generator, data extractor, SAP executor, and report writer, is the whole product. Frekil calls the underlying system a continuous, self-improving AI engine, and says every study a customer runs makes the next one sharper.
The problem it is aimed at
Clinical trials test a drug in a narrow, chosen group of patients under controlled conditions. Real-world evidence is the messier follow-up question: what happens once the drug reaches the far larger population the trial never enrolled, people with other prescriptions, other conditions, a different metabolism. Building that evidence means assembling a study out of data that was never collected for research, electronic health records built to document a visit, claims data built to bill an insurer. Turning either into a defensible study is slow, manual work, mostly out of public view.
Frekil states its own numbers for how slow: a single real-world evidence study can cost between $100,000 and $1 million and take 6 to 18 months, and most of that effort goes into cleaning, joining, and reshaping data before any analysis starts. Frekil’s bet is that the slowness is not the science, it is the plumbing, and that plumbing is what a language model paired with statistical code generation can automate.
The company draws a line around what the AI may touch. Its stated architecture keeps patient-level data walled off from the models: the AI generates statistical code in R, Python, or SAS, and that code runs in a sandboxed environment against the customer’s own data, inside Databricks, Snowflake, AWS, GCP, or Azure. The company frames this as an engineering boundary, not a policy choice.
Founders, not epidemiologists
Nikhil Tiwari and Shivesh Gupta met as students at IIT Bombay. Before Frekil, Tiwari wrote software at Stripe, Amazon, and Marsh McLennan, and spent a semester on exchange at the University of Geneva, near CERN, per a recruiting profile of his background. Gupta worked as a systems software engineer at Sony in Japan and, by his own account on Y Combinator’s site, turned down a full-time offer from a high-frequency trading firm to build Frekil instead. Neither founder came from pharma or biostatistics.
That outsider start shows up in how the company talks about itself. Its stated operating principles include a bias for action, with founders promising customer support responses within ten minutes around the clock, alongside scientific rigor with no methodological shortcuts. For a very small team selling into a cautious industry, a ten-minute reply time reads less like a perk and more like a way to compensate for not yet having a long track record.
From scanning images to generating evidence
Frekil did not start as a real-world evidence company. Its original Y Combinator launch post, titled “Scale AI for Medical Scans,” described a platform to procure and annotate medical imaging datasets, X-rays, CT scans, MRIs, for AI model training, built around radiology-chain partnerships and a marketplace of radiologists doing the annotation. An archived Crunchbase snapshot from April 2025 describes the same business, annotating images for healthcare AI teams, device makers, and pharmaceutical companies. Crunchbase’s own internal identifier for the listing, still visible in its web address, is built around the word “Notate,” suggesting an earlier working name.
By the time the company appears in later coverage and on its current website, the product is entirely different: not annotating images, but generating real-world evidence from structured clinical data. The pivot itself is undocumented in any source found for this profile; what exists is a before-and-after picture, an imaging-annotation startup in early 2025 and an evidence platform by the time Inc42 profiled it in mid-2026, with the same two founders connecting the two.
The company went through Y Combinator with partner Nicolas Dessaigne. YC’s own page lists the batch as Spring 2025, though a Forbes Business Council bio and a YesPress feature both call it the X25 batch, and Tiwari’s LinkedIn headline uses a third label. A small detail, but the sources disagree, and no single label appears everywhere.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| India registration | Frekil AI India Private Limited, CIN U62099KA2025FTC203413, incorporated May 28, 2025 with the Registrar of Companies, Bangalore. Filed as a subsidiary of a company incorporated outside India. Registered office in Basavanagudi, Bengaluru. Authorized capital of 1,00,000 rupees, paid-up capital 10,000 rupees, a minimal figure for a new subsidiary. Directors are Nikhil Tiwari and Shivesh Gupta. | Public record |
| US entity | Legal name given as Frekil Inc. on an archived Crunchbase page. No filing under that name, or any close variant, turned up in an SEC EDGAR search for this profile. | Not found |
| Y Combinator batch | YC’s own company page lists “Batch: Spring 2025” with Nicolas Dessaigne as primary partner. A Forbes Business Council bio and a YesPress feature both instead call it the X25 batch. | Differs across sources |
| Funding total | Tracxn and Standout.work put total funding at $500,000 from a seed round dated June 2025, sourced from Y Combinator. Prospeo lists total funding at $1,000,000. A YesPress feature says the round included Y Combinator and 7BC Venture Capital, without a figure. No SEC Form D turned up to settle it. | Differs across sources |
| Original product | Frekil’s own 2025 Y Combinator launch post describes a medical-image annotation platform, not a real-world evidence platform. An archived Crunchbase snapshot from April 2025 describes the same annotation business. | Public record |
| Named pharma customers | None found. The site publishes three unnamed user testimonials describing specific analyses, an ACE inhibitor versus ARB comparison, a sepsis prognostic score validation, and a treatment-sequence sensitivity analysis, without naming the individuals, employers, or drugs. | Company-stated |
| Published studies | No peer-reviewed papers or PubMed entries tied to the company were found. | Not found |
Read plainly: the one hard public record here is the Indian subsidiary’s registration, a minimal-capital entity typical of a young startup’s local office, not evidence of scale. Funding, batch naming, and customers all come from the company itself or from trackers that disagree, and the product Frekil sells today is not the one it launched with a year earlier.
What to watch
- Whether a Form D or other financial disclosure surfaces to confirm the reported seed amount and 7BC Venture Capital’s participation.
- Whether Frekil names a pharmaceutical, biotech, or CRO customer publicly, rather than the unnamed testimonials on its own site.
- Any peer-reviewed or conference-presented study that credits Frekil’s pipeline.
- Whether the founders ever explain the shift from medical-image annotation to real-world evidence generation.
In their words
“Today, it takes months to understand how drugs perform in the real world, delaying safety insights and leaving patients exposed to unknown risks and benefits for far too long.”
Frekil company description, Y Combinator and company site · Company-stated
“I’m the co-founder and CTO at Frekil, building AI-powered workflows to generate automated real world evidence of drug performance. Previously, I worked as a systems software engineer at Sony Japan and declined a full-time offer from an HFT to build Frekil.”
Shivesh Gupta, co-founder and CTO, Y Combinator founder bio · Company-stated
“Frekil is an end-to-end platform to procure and annotate medical imaging datasets like Xrays, CTs, MRIs, etc 10x faster with AI assistance.”
Nikhil Tiwari, 2025 Y Combinator launch post · Company-stated
“AI generates the code. The code runs in a sandbox. The data never leaves.”
Frekil company site, on its data-handling architecture · Company-stated
“I ran a target trial emulation to compare ACE inhibitors vs ARBs for patients with heart failure. The tool generated the exact statistical code I needed to handle confounding, and gave a transparent, auditable trace from the raw data to the final survival analysis.”
Unnamed user testimonial, Frekil company site · Company-stated
“Frekil is an AI-powered research platform that aims to cut this process from months to minutes.”
Inc42, “30 Startups To Watch,” Aug 2026 · Independent
“Frekil’s name is an argument in acronym form: Fast Real-world Evidence & Knowledge Insight Loop.”
YesPress feature on Frekil, 2026 · Independent
“Neither came out of pharma, which is either the obvious risk or the point, depending on how you read it.”
YesPress feature on Frekil, on the founders’ backgrounds, 2026 · Independent
Related companies
Sources
- Public recordMCA company record, FREKIL AI INDIA PRIVATE LIMITED, CIN U62099KA2025FTC203413
- Public recordMCA company record, FREKIL AI INDIA PRIVATE LIMITED, CIN U62099KA2025FTC203413
- YC dataFrekil: RWE Generation in minutes, company and founder page
- CompanyFrekil homepage
- IndependentFrekil organization page, archived snapshot describing the medical-imaging annotation business
- Independent30 Startups To Watch: Startups That Caught Our Eye In July 2026
- IndependentFrekil: Real-World Evidence at Software Speed
- CompanyNikhil Tiwari, CEO, Frekil, profile
- IndependentFrekil funding rounds and investors
- IndependentFrekil profile, funding and founder bios
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
