
Turning AI research into products

Integrating AI into Products and Teams ✨ — with Disheng Qiu (Translated)
Topic 4 of 6
Published Feb 9, 2024 · Clip 30:48–43:28 (12:40)
Disheng explains the challenges of combining researchers with product teams: different backgrounds, objectives (publications vs business KPIs like conversion rate), and work rhythms (long-term research vs sprints). At Translated there are two modes: product-driven, where the product team defines a clear AI need with thresholds (e.g., classification precision/recall) and a quarterly roadmap reviewed against those thresholds; and merged teams, as with ModernMT, where product and research work almost as a single group with engineers, AI experts, and data scientists. He illustrates the latter with the new Human-in-the-loop product, where an algorithm ranks and routes the worst machine-translated sentences to human review, continuously improving the engine with almost no customer input. He stresses keeping research and product close and avoiding silos, and that customer interaction should be managed by experienced product people since researchers speak a different language.