Dropbox uses a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → to produce training labels for Dash search, while a smaller ranking model orders results when users search. Human reviewers first score a limited set of non-sensitive internal examples. Those judgments calibrate the language model, which then labels much larger collections of query-document pairs. Dropbox's explanation describes this as an offline teaching process.
This arrangement matters because Dash can give its answer-generating model only a small selection of retrieved documents. Better relevance training helps determine which documents reach that stage, while the smaller ranker avoids the latency and context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → limits of asking a large language model to rank every candidate during a search.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
videoDistill the LLM, Don't Serve It: Search & Personalization at DoorDash — Raghav Saboo, DoorDashAI Engineer
articleHow we optimized Dash's relevance judge with DSPyIlya Yakovlev,Andrew Cheung,Binoy Dash,Simran Jumani,Dmitriy Meyerzon,Mark Breitenbach,Ishan Mishra,Kazuaki Okumura,Mike White,Kevin Altschuler,Facundo Agriel,Ishan Mishra,Eric Wang,Dmitriy Meyerzon
articleHow we used DSPy to turn AI evaluations into better responses in Dash chatIlya Yakovlev,Andrew Cheung,Binoy Dash,Simran Jumani,Dmitriy Meyerzon