Parallax: Observability for machine learning systems

Parallax shows machine learning teams what their models are doing in production. It tracks input data, predictions and outcomes, alerts engineers when accuracy starts to drift, and traces a bad prediction back to the feature or data source that caused it. For teams running large language models, it also monitors cost, latency and response quality per request.

Arjun and Hannah ran the model platform at a large online marketplace, where a silent failure in a pricing model once went unnoticed for nine days. Rafael led our investment in the Series A in 2022 and sits on the board.

Parallax now monitors more than 30,000 models for 850 companies, from fraud detection at banks to recommendation systems at retailers. Its annual recurring revenue has more than tripled in the past eighteen months, and the team has doubled to 160 people.