Alex Albert works at Anthropic on making Claude useful to developers and users. His public work includes prompt engineering, developer education and research communication. He previously created JailbreakChat, a collection of prompts that tested chatbot restrictions.[2][3][7]
Start with Introducing Contextual Retrieval →Background
Albert’s June 2023 résumé records a BA in computer science at the University of Washington, with attendance from September 2019 to December 2022. His internships included an internal YouTube Ads debugging interface at Google and a correspondence-management tool at Capital One. He also worked on robotics and student entrepreneurship programs. Before joining Anthropic, he built JailbreakChat, publishing a website and source code for collecting prompts that attempted to bypass chatbot restrictions.[3][5]
Career
Anthropic’s instructor biography identifies Albert as its first prompt engineer and describes later work across developer relations and product, including technical content, startup relationships and developer strategy. His personal site summarizes the goal as helping people do more with Claude. That combination places him between model capabilities and the practical questions developers face when adopting them: how to provide context, connect tools and tell whether a workflow is useful.[2][1]
Contributions
Anthropic credits Albert with helping shape its Contextual Retrieval article. The method adds a short explanation of where each document chunk belongs before creating search indexes. This can preserve a company name, reporting period or other context that disappears when text is split into fragments. The article combines semantic embeddings with keyword retrieval and discusses reranking. Its examples show why retrieving a plausible fragment is insufficient when the fragment has lost the details needed to answer a question.[6]
Tools, writing & talks
Writing, research & talks
Start here
Introducing Contextual Retrievalanthropic.comAn illustrated explanation of retaining document context when preparing chunks for retrieval.
Begin with the retrieval article’s worked example of a financial document chunk. Familiarity with text search helps; no model-training background is needed to understand the failure it addresses.
Credited with shaping the article.[6]- Building with MCP and the Claude APIyoutube.com
Anthropic API and MCP engineers explain the Model Context Protocol as a universal connector between models and external systems, and how to use MCP connections through the Claude API.
Catalog workOriginal source - Building more effective AI agentsyoutube.com
Anthropic's Alex and Erik discuss how RL on coding and search tasks makes Claude a capable agent, and how multi-agent setups act as test-time compute for better answers.
Catalog workOriginal source
Tools
- JailbreakChat
github.comSource for the site collecting prompts intended to probe chatbot restrictions.