Saad Aziz

Built production-grade multi-agent orchestration systems from scratch using LangGraph and LangChain to automate clinical workflows spanning SOAP note generation, drug interaction analysis, and patient communications. Implemented solutions emphasizing tenant isolation and robustness, including fixes for isolation bugs in a Weaviate multi-tenant setup via explicit tenant queries to prevent cross-tenant data leakage, and SHA-256 content hashing for deduplication. Added persistent chat history with user-level context management, real-time voice/websocket support for hands-free clinical documentation, and retrieval quality/embedding performance evaluation using LangSmith (including prompt versioning and monitoring). Delivered enterprise NLP-to-SQL capabilities for non-technical users to query 50+ EHR tables in a HIPAA-compliant way, with read-only query validation and sensitive column filtering to prevent data exposure. Implemented reflective error handling for SQL execution failures with

Built a low-latency, HIPAA-compliant intelligent doctor assistant with voice integration, supporting real-time daily interactions across text, voice, and WebSocket channels for 50+ physicians. Implemented a multi-tool agent connected to EHR data sources such as appointments, prescriptions, lab results, and patient records, with reflective tool-call recovery and automatic handling of transient API errors. Designed mandatory confirmation workflows for high-risk actions (appointment booking/cancellation) using template-based validation to reduce medical errors. Added temporal-aware prompt engineering for accurate appointment filtering and scheduling, and used LangSmith dashboards to monitor agent latency, tool-call success rates, and conversation quality in production.

Developed a reference-based and voice-interaction RAG chatbot that answers using source text chunks and provides document references across multiple file types. Implemented AWS Lambda triggers for automated processing of uploaded files to S3 for zero-touch document ingestion, and added voice interactions using WebSocket for hands-free usage. Built with LangGraph/LangChain and supporting infrastructure such as ChromaDB and FastAPI for retrieval and API delivery.