Glean Details How Zero-Shot System One Routing Runs Invisibly Inside Enterprise Search
Enterprise search leader Glean pairs Jev with frontier LLMs and traditional classifiers so fast-changing customer taxonomies no longer require constant retraining.
Among the earliest large-scale enterprise adopters to discuss its architecture publicly, enterprise search platform Glean shared how it embeds TypeSafe AI’s Jev inside live production routing flows alongside both traditional classifiers and full-scale LLMs.
Zero-shot flexibility for shifting enterprise taxonomies
Glean engineering leader Chinmay Goyal explained that traditional supervised classifiers work well when label definitions stay static, but break down quickly in enterprise search where customer workspaces, connector permissions, and routing taxonomies evolve continuously. Retraining and redeploying bespoke classifiers for every schema tweak slows product velocity.
By inserting Jev as a zero-shot System One layer, Glean adapts routing criteria immediately when a taxonomy changes, while still reserving heavyweight reasoning LLMs for open-ended synthesis and complex multi-hop answers.
Invisible infrastructure for end users
Goyal emphasized that enterprise customers neither know nor need to care whether a given query hop ran through Jev, a classical model, or a frontier LLM. Instead, Jev operates quietly inside the routing fabric—delivering fast, low-cost triage that keeps end-to-end search latency crisp while hiding orchestration complexity from the user.