← All questions
System DesignVery Hard· Latency· Auditability· LLM Orchestration

Your team wants every stage collapsed into one LLM call...

The auditability requirement does not necessitate separate round trips. By structuring the single LLM call to output a schema-validated record for each sub-process, you achieve the same level of logging and rollback capability as a five-call trace. Furthermore, it is worth questioning whether a 300 ms SLA is realistic for an LLM-driven query processor, regardless of the call structure.

Code playground (Python)