Etio

A financial reporting co-pilot for insurance regulatory closes, where deterministic code makes every number and the model supplies the language around them. By Ari Joury Project page: https://towardsdatascience.com/shipai/projects/etio ShipAI is a Towards Data Science project. You can learn more about it here: https://towardsdatascience.com/shipai/about Key takeaways. 1. A deterministic renderer fills text slots from the ledger and aborts the whole render if any slot fails to bind. 2. Triage clusters findings by walking fact lineage with graph traversal, not a model, surfacing one root cause over 20 symptoms. 3. The verifier is attacked in CI across seven error classes, each locked to a baseline of zero misses. Chapters: 0:00 Numbers flow facts to prose 1:08 Explainable triage 3:49 Commentary built from slots 5:04 Catching an edited number 6:44 Grounded chat and refusals 8:48 Attacking the verifier 10:29 Causal discovery and scenarios 12:29 Six use cases, one rule