Preston Zhou is Relace’s co-founder and chief executive. He moved from physics research into building models and infrastructure for coding agents. Relace now also provides inference for open-weight models.[1][2][3]
Start with Relace’s models and infrastructure for coding agents →
Career
Zhou’s founder biography describes a move from physics to machine learning. Harvard’s Mitrano Lab lists him among its former graduate students. He co-founded Relace with Eitan Borgnia, developing infrastructure around the operations coding agents repeat: finding context, applying edits and managing repositories.[2][5][4]
Contributions
Separate code editing from general reasoning
In their joint 2025 account, Zhou and Borgnia describe specialized models for embedding, reranking and applying edits. Separate tools locate relevant code and merge proposed changes. Their announced repository service extends this approach to source-control infrastructure.[1]
Expand into open-weight inference
Relace’s current homepage describes a broader inference service and names both founders as contacts. Its stated focus is improving utilization and efficiency in model serving as demand grows. That extends the company’s earlier specialization in coding-agent models; it should not be confused with Zhou personally authoring each model the service hosts.[3]
Tools, writing & talks
Writing, research & talks
Start here
The founders’ explanation of embedding, reranking, edit application and repository infrastructure.
Start with the infrastructure section and distinguish retrieval from applying an edit. Basic software-development knowledge helps; performance figures are company-reported, and the current service has expanded beyond this 2025 account.
Co-author[1]