Mohammad Sadr

Designed and developed scalable FastAPI microservices for AI-powered voice assistants, conversational AI, and CRM platforms. Built LLM-based systems using RAG, embeddings, tool/function calling, and LangGraph multi-agent architectures for intelligent automation and tool integration. Engineered event-driven architectures with pub/sub and queueing for low-latency, high-throughput AI workloads. Orchestrated asynchronous workflows and real-time pipelines using Temporal and Debezium (CDC). Developed speaker diarization and meeting intelligence for structured summaries, decisions, and action items. Applied DSPy for prompt optimization and evaluation and implemented Pydantic-AI schema-driven systems for structured outputs and validation. Implemented LLM observability and evaluation using LangSmith and Langfuse to monitor quality, latency, and hallucination rates. Standardized AI-assisted development workflows and led multi-cloud deployments using Terraform across Azure and GCP.

Built end-to-end ML infrastructure using Kubernetes, Docker, and GPU clusters for scalable training and inference. Fine-tuned LLMs with Hugging Face, PEFT, and LoRA to improve domain-specific accuracy. Developed RAG systems with LangChain, embeddings, pgvector, and OCR for real-time enterprise search (<1s latency). Designed agent systems using LangGraph and MCP for multi-agent orchestration and tool-based reasoning. Led a team of 4 engineers, improving delivery speed by improving workflows and code practices.

Designed scalable AI/ML platforms using TensorFlow, PyTorch, and scikit-learn. Built automated infrastructure using Docker, Kubernetes, and Ansible, reducing setup time by 80%. Improved reliability and reproducibility of ML pipelines across distributed environments.