§ feed · storyline
Photon: Building a Retrieval and Ranking Engine from Scratch
Perplexity releases Photon, an in-house retrieval and ranking engine that cuts p99 latency from 800 ms to 65 ms and reduces estimated cost per agent task by 68%.
Perplexity described Photon, an in-house retrieval and ranking engine built for its search workloads. The migration reportedly reduced p99 retrieval-and-ranking latency from about 800 ms to 65 ms, while a faster Search API preset offered 160 ms p50 latency and reduced estimated model-plus-search cost per agent task by 68% in evaluated benchmarks.
§ sources1 publication · timeline below