DeepL: A Product Culture that Creates Top-Notch AI Products
summary
DeepL runs its own data centers, trains in-house models, and supports 29 languages with Pro availability in the EU, US, Japan and Korea; product team is ~30 people from 14 nationalities. Workflows use ClickHouse, Metabase and Grafana for tracing, a unified experimentation platform for A/A and randomized 1%→10% rollouts, and strict GDPR rules that exclude using Pro data to train models. Teams follow product triad staffing, Jobs‑to‑be‑Done segmentation and Team Topologies to decide what to build next.
AI-written summary · about
chapters (19)
- 0:00Introduction to DeepL
- 1:07What is DeepL
- 3:37Product management at DeepL
- 6:27Remote-first culture at DeepL
- 8:14Pillars of innovation at DeepL
- 10:19Product team structure
- 13:47Collaboration with research
- 15:34Data-driven approach
- 17:19Product story: DeepL Write
- 19:06Experimentation and rollouts
- 20:11Beyond current customers
- 23:00Jobs to be done framework
- 24:24Flywheel model
- 25:07Product team learning
- 26:31Why customers pay
- 27:34Cross-functional teams
- 29:42DeepL’s success factors
- 31:07Challenges at DeepL
- 32:48Q&A on remote work
related talks
30mBuilding an AI-Powered Organization
28mHow to Compete with OpenAI and Google
40mHow to build, manage and retain a killer team in Deep Learning
19mComputers are still dumb: bringing your AI magic to enterprises
18mSmall AI Teams with Huge Impact
32mEncoding Company Culture in Applied AI Systems
45mDeepBrew – AI
11mUsing AI to extend the range of data scientist