Abstract
The present disclosure relates to an end-to-end agentic artificial intelligence system capable of autonomously ingesting live serving logs, isolating performance issues, mathematically verifying hypotheses using deterministic tool-calls, and deploying safe, low-latency fixes to a serving stack. The system chunks data into semantic problem slices and generates natural-language hypotheses. To prevent hallucination, the system executes deterministic queries against live data to verify its own hypotheses. Verified insights are translated into simple, non-machine-learning programmable rules that execute rapidly in production. These rules are pushed to a limited fraction of live traffic, where a real-time monitor compares performance against a baseline, automatically shutting down underperforming actions and feeding logs back into the system to complete a continuous learning loop. Keywords: Agentic AI, log analysis, automated remediation, content serving, hypothesis verification, large language models, fractional serving.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Ramamurthi, Indu; Lagisetty, Raghav; Joshi, Ajay; Narayanan, Srinath; Singh, Sarvjeet; and Narayana, Pradyumna, "System and Method for End-to-End Agentic Log Analysis, Verification, and Automated Remediation in Content Serving", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10174