SYSTEM Cited by 1 source
GEPA prompt optimizer¶
GEPA is a prompt-optimisation method published in arXiv 2507.19457 and referenced by Databricks Genie as the technique used to optimise per-sub-agent prompts in its Multi-LLM architecture. The 2026-05-08 Databricks engineering post on Genie cites GEPA as the method enabling "the corresponding accuracy and cost can be further optimized using methods like GEPA" on table-search sub-agents.
Stub page — first wiki ingest naming GEPA as a tool referenced by a production system.
What's disclosed¶
From the 2026-05-08 Databricks Genie post:
- GEPA is referenced as a prompt-optimisation method applied to sub-agent prompts in Genie's Multi-LLM architecture.
- GEPA is the enabler for the simultaneous accuracy + cost gains in the "Multi-LLM" architectural advance: different sub-agents use different LLMs with optimised prompts (the "with optimised prompts" clause is what GEPA does).
- The disclosed application is table search — "different LLMs perform on table search tasks and how the corresponding accuracy and cost can be further optimized using methods like GEPA" (Figure 6).
- Method details are not reproduced in the Databricks post; readers are referred to the arXiv paper.
What's not disclosed¶
- How Genie integrates GEPA into its prompt-management plane (build- time vs runtime, periodic re-optimisation cadence, feedback loop shape).
- Which specific sub-agent prompts in Genie are GEPA-optimised vs hand-tuned.
- Whether GEPA is run per LLM (different optimised prompts per candidate model) or once across the model portfolio.
Seen in¶
- sources/2026-05-08-databricks-pushing-the-frontier-for-data-agents-with-genie — canonical first wiki disclosure of GEPA being used by a production data agent. Cited as the prompt-optimisation method in the Multi-LLM architectural advance: enables accuracy + cost gains simultaneously on Genie's table-search sub-agents (Figure 6).