Samir Unni

@DOGE_HHS published a dataset with billing activity for every Medicaid provider; 227 million rows, from 2018 to 2024. I cross-referenced it against the federal NPPES provider registry and @OIGatHHS's exclusion list, and started looking for fraud. It didn't take long to find providers banned for fraud still collecting payments a decade later. Brand-new entities billing $170 million in their first 18 months. A lab that went from 73 claims to 48,000 in a single month. Here are five cases: 1. We Care Transportation, a Kansas transport company, was excluded from federal healthcare programs in January 2010. Their NPI was never deactivated. They kept billing Medicaid for another ten years, collecting $4.4 million through July 2020. As of today, their NPI is still active and their exclusion has not been lifted. They're one of 13 excluded providers in the dataset who continued billing after being banned -- $7.3 million in payments that were improper by definition. 2. Fishing Point Health Care LLC registered in Virginia as a "General Practice" in February 2023. Eighteen months later it had billed Medicaid $170 million, almost all of it on a single attendant care code (S5121). Claims went from 1,213 per month to over 10,000. Toward the end of 2024, it started tacking on ancillary codes -- labs, therapy, E&M visits -- which looks a lot like an attempt to make the billing profile seem more legitimate. There are 20 other entities formed since 2022 that have each billed Medicaid more than $50 million, totaling $1.7 billion. 3. Lifeline Biosciences LLC, an Illinois lab formed in early 2022, has exactly four rows in the entire dataset. In August 2023 it billed 73 claims for infectious agent detection. In December, it billed 48,355 -- a 662x increase in one month, good for $7.3 million. The national trend for that same code was flat. Forty-three providers had similar one-month explosions (10x or more, over $1 million) in the post-COVID period, across 48 separate events. 4. New Life Wellness Center LLC in Phoenix was excluded in November 2024 after a fraud conviction. The billing history tells you everything: 25 patients and $280,000 a month in mid-2021, ramping to 273 patients and $8.7 million a month by March 2023 across 14 behavioral health codes. Community mental health claims alone went from 125 to 5,603 in five months, then billing stopped abruptly. More than 4,200 behavioral health providers in the dataset have a similar growth curve -- peak monthly claims at least five times their trough, with total billing above $5 million -- a $98 billion pool that could use a closer look. 5. C&C Mental and Family Services, a Florida entity formed in April 2023, billed $3.6 million on a single code: H2019, therapeutic behavioral services. The monthly volumes make no sense -- 1,180 claims in October 2023, 804 in February 2024, then 34,760 in July. That's a 43x swing against a flat national trend, which is not what a real clinic's patient load looks like. Over 1,200 providers on behavioral health, attendant care, or waiver codes have the same kind of volatile billing while topping $1 million in total payments, accounting for $14.9 billion collectively.

Wonderful contrast between AI at its best (we help you) and AI at its worst (we take away your humanity)

Say it five times fast: zero pricing power

So…does this mean every app will start to look like WeChat?

Yesterday's news, that @AnthropicAI acquired @CoefficientBio for $400M drew attention mostly for the price tag and the stark contrast with @OpenAI's acquisition of @tbpn. But far more interesting is what the deal reveals about how @_DimensionCap, the venture firm behind Coefficient, actually operates. Coefficient looks less like a traditional portfolio company than an incubated spinout: Dimension effectively cofounded, and owned half, of Coefficient, having seeded it with their internal infrastructure and workflows. That makes a lot of sense when you look at Dimension's founding team: a scientist and company incubator (@AGoulburn), a computational biology investor (@nanli), and an AI/ML-native technologist (@zavaindar). Their backgrounds read less like a conventional VC partnership and more like the founding team of a technical platform. The much-discussed "Dimension-GPT", while not a public product, sounds like an internal AI operating system for scientific judgment and decision-making that eventually became powerful enough to spin out as a company in its own right. What that OS likely looked like in practice is worth spelling out, because it suggests a replicable model: 1. Ingest external scientific and commercial literature 2. Normalize it into a proprietary knowledge layer 3. Run LLM-driven reasoning workflows to map claims against evidence and market structure 4. Call specialized bio/ML tools where needed 5. Produce memos, rankings, and opportunity maps that compound over time 6. Eventually, add agentic loops that propose their own follow-on analyses. This isn't "AI for biology" as a buzzword, but AI as the connective tissue between technical underwriting, company formation, and portfolio construction. Most venture firms treat AI as a sector thesis, but Dimension's model suggests a few may be quietly building domain-specific AI systems that function as company factories. If that pattern scales, the most consequential AI applications in life sciences may not come from startups pitching VCs — they may come from the VCs themselves.

Live look at @nabeelqu and @WillManidis working on their upcoming book…

Acquihires are the startup world equivalent of collective bargaining

Yesterday, I cross-referenced suspicious Medicaid billing with federal campaign donations, to find multiple flagged entities donating to senators on the Finance Committee, which oversees Medicaid But Congress isn't where the only action is: Medicaid's administered by states. The single biggest finding: two California community health centers flagged for extreme billing spikes donated a combined $1.23 MILLION to a committee fighting a state ballot measure to bring transparency to those very community health centers! In VA, the Upper Mattaponi Indian Tribe billed Medicaid $57M in just 2024, during which they made $55K+ in donations, incl $25K to the new governor's inaugural committee. And the CEO of Wall Residences ($315M in Medicaid billing, 24x monthly spikes) made 40+ donations in VA, from 2019-2024. In TN, D&S Residential Services (flagged for 11.4x billing spikes) operated a PAC that gave $6.5K to the state Comptroller—the official who audits state spending. I don't need to explain why a Medicaid provider donating to the person who audits Medicaid is worth a closer look. NJ had the most interesting structural pattern. 4 of 5 entities flagged for dual-billing BH + transport codes also showed up in state campaign finance records. Donors targeted legislators with health committee jurisdiction or who represent the exact districts the entities operate. Once again: none of this proves fraud or corruption. Campaign donations are legal. But here's the reality that we can't ignore: entities with suspicious Medicaid billing are donating to the same elected officials who oversee Medicaid

The single biggest finding: two California community health centers flagged for extreme billing spikes donated a combined $1.23 MILLION to a committee fighting a state ballot measure to bring transparency to those very community health centers!

There’s Medicaid and private insurers, but also Medicare. @DOGE_HHS could release the same data for Medicare, where there’s nearly 2x as much fraud as Medicaid!

A major reason we have so much healthcare fraud is because of how Obamacare works: the less fraud there is, the less profit the health insurers make. No wonder they don't care about fixing this!

I went through the different categories of potential fraud in my previous post, and dug out dozens more providers with suspicious billing. The total amount of potential improper payments, across just these providers, is $3.5B!

Overheard during a tech company Super Bowl debrief: "is the quarterback like a product manager?"

Tried setting this up, but can’t get PowerPoint to recognize the Microsoft 365 account in which I installed the Claude add-on. Once the issue of handling multiple M365 accounts is resolved, we’ll be within spitting distance of AGI.