• Ayush Kumar Choudhary
• Dr. Y. Sarojini
• Jyotsna G
• Ashish Sharma
• Aptha K. S.

Large language models (LLMs) enable automated program repair, but patches that pass available tests may still be semantically incorrect. Overmend is an autonomous software recovery system that combines runtime failure ingestion, AST-based fault localization, LLM patch synthesis, isolated sandbox validation, mutation-based trust evaluation, and policy-aware GitHub orchestration. After a candidate passes the available tests, controlled AST mutations are applied only to the changed region; test sensitivity is incorporated into a deterministic trust score together with patch size, blast radius, and policy constraints. The implementation was validated with 150 automated tests, while a separate evaluation on 20 Python bugs compared tests-only and tests-plus-trust acceptance. Tests-only accepted four incorrect patches (20% observed false acceptance); trust evaluation blocked all four in the evaluated sample. Fault localization required 80–150 ms, mutation analysis 6–15 s, and the reported detection-to-Pull-Request path 15–35 s. These results support mutation evidence as an additional acceptance signal, while the limited benchmark and Python-only scope preclude claims of general correctness.

automated program repair, autonomous software recovery, large language models, patch synthesis, mutation testing, fault localization, trust scoring, software reliability, Docker sandboxing, GitHub automation.

"Overmend: AI-Powered Autonomous Recovery System Using Patch Synthesis and Mutation-Based Trust Evaluation ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.c57-c64, October-2026, Available :http://www.jetir.org/papers/JETIR2610209.pdf

"Overmend: AI-Powered Autonomous Recovery System Using Patch Synthesis and Mutation-Based Trust Evaluation ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppc57-c64, October-2026, Available at : http://www.jetir.org/papers/JETIR2610209.pdf