
A comprehensive distributed system for detecting vulnerabilities in project dependencies, published in the 16th International Conference on Computer, Communication and Network Technologies (ICCCNT), IIT Indore, India, July 2025.

Developed a PPO-based reinforcement learning pipeline to fine-tune LLaMA models, enhancing human-like text generation to bypass AI detection while preserving semantic coherence using RLHF with reward normalization and KL-divergence constraints.