Building the system of record
for how claims get decided.
Insurance claims adjudication is one of the largest under-automated processes in healthcare. Trillions of dollars and lakhs of crores move through it every year, decided by humans reading PDFs at 12 minutes per claim. We're changing that — without losing the cited reasoning and signed audit trail regulators require.
Two markets
US payers and Indian TPAs. Different regulators, different price points, same core problem.
Two offices
San Francisco for US sales and ML research. Mumbai for the engineering majority and Indian go-to-market.
Small team, high bar
We hire generalists who can ship. Forward deployment, infra, security, backend, ML — all five disciplines staffed before sales hires.
Why now
Three things changed in the last 18 months. Open-weight LLMs became good enough to fine-tune for a specialist task and beat GPT-4-class general models on it. Schema-constrained decoding made structured-JSON output reliable. And inference economics fell by an order of magnitude.
Together, these unlock something that was technically impossible before: a model you can fine-tune on real adjudication data, constrain to your decision schema, run inside a customer's VPC, and audit cryptographically. That's Adjudo.
Operating principles
We train, test, and ship on the same kind of claim a TPA officer rejects on a Tuesday afternoon. Synthetic-only is theatre.
Every decision references the clauses it relied on. The model loses points if it skips a citation, even when right. Reviewers shouldn't have to take our word for anything.
Every adjudication writes a signed, append-only audit row. Tampering breaks the chain. Compliance isn't a feature — it's the foundation.
Some customers are cloud-native US payers. Some are NABH-bound Indian hospital chains. We bring the same product to both.
Want to help build it?
We're hiring across forward deployment, infra, security, backend, and AI/ML — in San Francisco and Mumbai.