Srinidhi Moodalagiri

entrepreneur

[Srinidhi Moodalagiri @smoodsiee]

Single-tenant models cost more to deploy, but as of now, I think they're the only architecture that actually aligns with how accountability works in this industry. Any benefits that shared infrastructure may bring don't matter if they make the output legally indefensible.

[Srinidhi Moodalagiri @smoodsiee]

Multi-tenant architectures are great for industries like SaaS. One model serves many customers, and costs get driven down. Healthcare companies are only just starting to realize that the same doesn't apply to them.

Coverage policies vary wildly across payers in scope and logic.

[Srinidhi Moodalagiri @smoodsiee]

If a model makes a decision for Plan A, and that decision gets audited, the payer needs to prove the model was trained and tuned specifically for their policies. If the model is shared across 50 plans, that proof becomes almost impossible to confirm.

[Srinidhi Moodalagiri @smoodsiee]

It doesn't make sense for stable, well-managed patients to be constantly struggling.

Yet the exact population it's supposed to be serving is being penalized time and time again.

[Srinidhi Moodalagiri @smoodsiee]

39% of people with chronic conditions cited prior authorization as their single largest burden, which is sadly unsurprising.

Chronic condition management is predictable. https://t.co/WDlJolxOCJ

[Srinidhi Moodalagiri @smoodsiee]

In my eyes, chronic conditions are exactly where payers should have the most confidence. There's historical data, treatment adherence patterns, and outcome tracking. Prior authorization is meant to catch inappropriate utilization.

[Srinidhi Moodalagiri @smoodsiee]

As of now, we're leaving that feedback untouched.

[Srinidhi Moodalagiri @smoodsiee]

Appeals tend to get framed as an issue with patient rights or a regulatory safeguard. I think that framing is a bit shallow.

If you’re building AI-driven utilization management systems, appeals are one of the clearest signs you get about data and decision quality.

[Srinidhi Moodalagiri @smoodsiee]

They show exactly where policy logic, clinical nuance, and operational shortcuts have backfired.

And we throw this information away over and over again.

If we stop treating appeals as purely adversarial processes, we could dramatically reduce the need for them at all.

[Srinidhi Moodalagiri @smoodsiee]

There really aren't any perfect options. Healthcare data just isn't organized for machine reasoning.

[Srinidhi Moodalagiri @smoodsiee]

LLMs have context windows measured in tokens, and most healthcare applications hit those limits faster than people realize.

A prior authorization request holds a lot of weight. It's a clinical note, treatment history, coverage policy document, and sometimes imaging results

[Srinidhi Moodalagiri @smoodsiee]

But that only works if you've already structured your policy documents and clinical data in a way that supports precise retrieval. Let's be honest, most payers aren't doing that. They're still working with PDFs and unstructured text.