Technical writing from the team building Reductohttps://reducto.ai/https://reducto.ai/engineering/posts/speculative-decoding-throughput-latency/https://reducto.ai/engineering/posts/speculative-decoding-throughput-latency/We benchmark multi-token prediction across workloads and concurrency levels to show when accepted speculative tokens improve throughput—and when verification overhead takes over.Tue, 28 Jul 2026 00:00:00 GMThttps://reducto.ai/engineering/posts/llm-judges-table-ocr-false-positives/https://reducto.ai/engineering/posts/llm-judges-table-ocr-false-positives/We built a benchmark where ground truth is known by construction, then used it to show that frontier LLM judges hallucinate table OCR errors at non-trivial rates.Fri, 10 Jul 2026 00:00:00 GMThttps://reducto.ai/engineering/posts/streaq-scheduler-latency/https://reducto.ai/engineering/posts/streaq-scheduler-latency/Bursty, latency-sensitive traffic exposed the limits of per-tenant overflow pools. So we vendored a small task queue and built our own scheduling and traffic-management layer on top, keeping shared capacity warm, fair, and policy-aware while cutting parsing latency about 3x.Thu, 18 Jun 2026 00:00:00 GMThttps://reducto.ai/engineering/posts/fixing-cv-serving-bottleneck/https://reducto.ai/engineering/posts/fixing-cv-serving-bottleneck/Customer bursts exposed a weakness in our CV serving stack that average traffic mostly hid: the bottleneck was not model compute, but the pipeline around it.Wed, 03 Jun 2026 00:00:00 GMThttps://reducto.ai/engineering/posts/agent-native-observability/https://reducto.ai/engineering/posts/agent-native-observability/Most AI-for-observability demos start with a chatbot over logs. We cared about a less flashy question: could an agent understand our production system well enough to help operate it?Tue, 27 May 2025 00:00:00 GMT