At Kodamai, I architected multimodal clinical and document-intelligence RAG systems, including one that combines vector search, graph data, and SQL to query clinical records. I also developed an invoice–purchase order automation platform that processed 100K+ purchase orders for a major Saudi industrial client.
At micro1, I fine-tuned the live Zara AI interviewer, integrating speech recognition, speech synthesis, and semantic scoring to reduce inference latency by 40%. I also improved data preparation time by 50% and established an MLflow-based MLOps lifecycle that reduced deployment time by 60%.
At Pareto AI, I design evaluation tasks and programmatic verifiers for AI agents, and develop MLflow-tracked training pipelines. Independently, I engineered OmniRAG, a multi-capability AI platform with document Q&A, a voice agent, and a Trust & Safety moderation pipeline.
