Abstract
Autonomous network-assurance agents built on large language models must translate an operator's natural-language question into schema-valid device queries across hundreds of data-model modules. Conventional pipelines select the target schema by embedding-based retrieval, which is slow and non-deterministic, and ask the model to synthesize the protocol payload, which invites malformed or invented queries. An engine resolves the common-case query to its target modules deterministically from a precomputed lexical-anchor index, classifies the outcome into four confidence modes with abstain-to-retrieval as the fail-safe, broadens the read set with ground-truth-derived companion targets under an add-only superset-safety invariant that cannot alter the routing decision, synthesizes the concrete protocol operation by walking the compiled schema rather than by model generation, and gates dispatch so that reasoning-free model calls are removed. A capture-and-regression subsystem keeps the catalog correct as schemas evolve. The result is lower latency, fewer model tokens, and elimination of payload hallucination.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Deterministic Resolution, Amplify-Never-Create Expansion, and Schema-Directed Operation Generation for Network Management", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12687