Vesyra Studios builds working applied-AI systems and opens up every layer — the retrieval, the scoring, the model reasoning, and the audit trail. Not toy notebooks: deployed pipelines you can click through and interrogate.
Most AI demos show you an answer. These show you why — every claim traced back to the evidence that produced it.
Each project is a deployed application with a real database, real retrieval and a real model in the loop. You interact with the running system, not a screenshot of one.
Decisions carry their reasoning: matched criteria, missing criteria, confidence scores and the source chunk IDs behind every statement. Nothing is asserted without provenance.
De-identification, deterministic rule layers and full traceability are treated as first-class requirements — the constraints that decide whether AI actually ships in regulated domains.
A complete explainable-AI pipeline for US healthcare prior authorization.
Prior authorization is the slowest stage of the US revenue cycle: a reviewer reads a patient chart, opens a 40–100 page payer policy, and checks every criterion by hand. This system automates that comparison without losing the audit trail — hybrid vector retrieval over patient notes and payer policies, a deterministic rule engine scoring twelve clinical features, and a grounded LLM explanation where every claim cites a retrieved chunk.
Open the project