Engagement briefs.
We publish what an engagement involved and how outcomes were measured, with client names and numbers withheld. Full references are available under NDA during sales conversations, which keeps this page honest and those calls concrete.
Pre-procurement model evaluation
A retail group was choosing between two frontier models for customer service. The lab ran both against the client's real ticket workload on Cognion-style trials. One model's marketing advantage disappeared on their data; the contract went to the other, with an evaluation report the client now reuses for every renewal.
Ecommerce personalization rebuild
An online retailer's recommendation engine had drifted after a catalog migration. The lab measured the baseline first (conversion per session, coverage, latency), rebuilt the embedding pipeline, and shipped behind an A/B gate. The engagement ended with the client owning the measurement, not just the model.
Automation with an emergency brake
A support organization wanted LLM-drafted responses. The lab defined acceptance thresholds and a sampling plan before launch, deployed with a rollback path, and reviewed output weekly. Two prompt regressions were caught by the plan, not by customers.
Site resurrection after a failed migration
A publisher lost a decade of articles to a botched platform move. Restorix rebuilt the archive from Wayback Machine snapshots, verified internal links and assets, and returned the section to service with a redirect map for search engines.
The next brief could have your problem in it.
Every engagement starts with measurement on what you already run. Describe the situation; we will tell you what we would measure first, free of charge, in the first reply.