Applied Compute accelerates large-scale code search via specialized AI post-training
Partnering with Turbopuffer, the team trains a 35B parameter model to query precomputed indexes, achieving up to 100x lower costs and significantly reduced latency compared to frontier models.