Hemanth Raju

I’m an AI Model Trainer / LLM Evaluator working with structured RLHF feedback loops to evaluate 100+ LLM outputs across reasoning, coding, and instruction-following. I use prompt analysis, comparative response ranking, and error taxonomy tagging to turn qualitative issues into actionable signals that guide fine-tuning priorities and improve output quality across model iterations.

I also build end-to-end AI systems that emphasize measurable performance and interpretability. From an LLM + knowledge graph interviewer selection pipeline (LLaMa 3.1-70B + Neo4j, achieving 64% job-skill coverage) to XWF-Agg explainability-weighted federated learning for intrusion detection (99.27% accuracy while preserving full data privacy), I’m focused on practical deployment and clear decision-making. I’ve shipped computer vision workflows as well—like fine-tuning YOLOv9 for underwater plastic surveillance—while designing inference pipelines that are plug-and-play for edge monitoring.