At UID, I led the design, development, and deployment of AI systems for video-based authentication, document understanding, biometric verification, and speech processing. These included video liveness detection, ID card OCR, head action recognition, lip reading, and automatic speech recognition.
I built and maintained machine learning infrastructure with Docker-based services, model APIs, and an MLflow ecosystem for experiment tracking and model lifecycle management. A modular redesign and model optimization reduced runtime by up to 50% while improving stability and mitigating memory-related issues.
My projects include a Persian lip-reading system with an end-to-end data-generation pipeline, and a Persian ASR system fine-tuned with MMS and Whisper. I also built an enterprise agentic RAG platform that coordinates specialized agents across hybrid search, reranking, and multi-hop retrieval.
More recently, I developed conversational and multi-agent AI systems, including a customer support agent with multi-turn memory, tool calling, and human escalation. I’m currently focused on advancing my expertise in LLMs, multimodal AI, and agentic systems that bridge vision and language.

