Pamela Fox is a Principal Cloud Advocate at Microsoft helping developers build Python applications with Azure, Microsoft Foundry, GitHub, and Visual Studio Code. Her career spans Google developer relations, early engineering roles at Coursera and Woebot, Khan Academy programming education, and teaching computer science at UC Berkeley.
From developer advocacy to computing education
Fox worked on the Google Maps and Google Wave APIs in Mountain View and Australia before becoming Coursera’s second frontend engineer. At Khan Academy, she created computing curriculum and improved its live-coding environment. She founded the Girl Develop It San Francisco chapter, developed openly licensed instructional materials, taught courses including CS61A and CS302 at Berkeley, and helped start Woebot’s engineering team. Her professional biography connects that experience to her current work making Python and cloud development more accessible.
Hybrid retrieval with semantic reranking. Fox combines keyword search, vector retrieval, rank fusion, and semantic ranking because embeddings can introduce irrelevant results while missing exact terms. In her comparison of retrieval strategies, vector-only retrieval scored 2.79 out of 5 for groundedness, compared with 4.87 for text-only search and 4.89 for hybrid retrieval with semantic ranking. Because her synthetic dataset favored lexical overlap, she emphasizes evaluating the actual workload.
Evaluation beyond sample questions. Her ai-rag-chat-evaluator combines model-graded groundedness and relevance with programmatic citation checks. Fox has found that changing retrieval parameters or models can improve results more than prompt tinkering, while model-based evaluations introduce meaningful latency and cost.
Repeatable, production-minded Python architecture. Her AI Engineer workshop connects asynchronous Python services, response streaming, PostgreSQL, Azure AI Search, infrastructure as code, and deployment through Azure Developer CLI. She advocates token-based document chunking, particularly for multilingual content, and treats traceable citations, observability, and identity-based authentication as application requirements.
Build and deploy a Python chat app, add retrieval over database rows or documents, and follow the engineering decisions that turn a working demo into a testable application.
Getting an application running before building everything yourself
Build and deploy a Python chat application, then extend the same workflow to product retrieval and document answers with citations, query rewriting and evaluated search.
▶ Watch ↗