Chinchu Thomas

At HP Inc., I architected AI-native, persistent agentic systems that support dynamic skill loading, RAG, and memory/context management, including tool use for multi-step execution. I implemented governance for reliable agent workflows using stage-level human quality gates and feedback loops. I also built an agent observability and evaluation framework using OpenTelemetry, MLflow, LangFuse, and DeepEval, and led security/threat-modeling activities for enterprise agent deployments. Additionally, I worked on efficient and edge AI approaches such as model distillation, quantization, ONNX-based deployment paths, and synthetic data pipelines.

At Idiap Research Institute, I studied speaker differences and behavioral characteristics in informal learning using multimodal data (video, audio, transcripts, and metadata). I developed multimodal machine learning and unsupervised domain adaptation approaches, including techniques based on knowledge distillation, to improve speaker-attribute prediction. The research contributed to peer-reviewed publications.