Chau Tran

Chau Tran is a software engineer at Glean leading technical work on Glean Assistant and enterprise semantic search. They build workplace AI that can retrieve trustworthy company information, respect access controls, and apply an organization’s established practices to unfamiliar tasks.

Tran studied computer science at Brown University, held technical roles at Quora, and worked as a research engineer at Meta’s Fundamental AI Research organization before joining Glean. By 2023, they were helping lead its work on large language models and vector search, combining vector retrieval, keyword search, and personalization to connect assistants with relevant enterprise knowledge.

Their approach to enterprise AI assistants centers on permission-aware retrieval: models need current organizational information without exposing documents users cannot access. Continually retraining models on corporate data makes freshness, access control, explainability, and factual reliability harder to maintain.

• Workflows as agent training and evaluation. Tran treats an agent’s completed execution as a generated workflow. Established procedures can support step-by-step evaluation and training, while successful agent runs can surface reusable practices an organization had never documented.

• Workflow Search for organizational know-how. Their approach to enterprise-aware agents retrieves successful procedures at runtime, allowing models to reproduce familiar workflows, combine relevant examples, or improvise when no precedent fits. Company-specific practices remain essential even as underlying models improve.

• Authority beyond textual similarity. Enterprise search must distinguish trusted workflows from superficially similar alternatives. Tran emphasizes knowledge-graph signals including working relationships, prior success, and organizational context when ranking established procedures.

• Context management for long-running agents. Tran coauthored work on distributed agent context management, combining sandboxed tool execution, isolated subagents, conversation compaction, and search-first skill discovery. They also contributed to enterprise-grounded spreadsheets and refreshable interactive artifacts, extending organizational knowledge into shareable work products.