Cut Translation Costs by 50× for Millions of Product Sheets…
summary
A team built and industrialized a large-scale product-description translation pipeline using LLMs to translate millions of product pages into 10+ languages while cutting costs dramatically. They replaced expensive SaaS translation with LLM-based models (proprietary and open-source), ran cost/quality/latency benchmarks to choose models, and automated quality monitoring by using a second LLM as a judge instead of human annotation.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:50Problem context
- 6:07Model selection
- 11:53Translation evaluation
- 15:27MVP results
- 19:40Production deployment
- 20:28LLM tips
- 27:58Open-source models
- 32:56Model fine-tuning
- 39:03Conclusion
- 40:07Q&A
related talks
57mMachine Translation of Human Languages in the Age of LLMs: Is This the End of the Language Barrier?
18mBuilding an AI service without training – AI translations with LLMs
49mImproving Accessibility through Leveraging Large Language Models (LLMs)
40mLLMs - application and impact on business
31mLeveraging LLMs to build supervised datasets suitable for smaller models
26mAI-powered Productivity using Open-Source LLMs
28mHow to Compete with OpenAI and Google
33mDesigning LLM Applications for Massive Datasets