Multilingual Apps on LLM APIs
Serve users in many languages with consistent quality, sensible cost and testing per language.
A taste of a lesson
Users write in Portuguese, but my bot sometimes answers in English. Why?
Usually the context pulls it toward English: an English system prompt, English retrieved passages or English examples. Add an explicit instruction such as: always reply in the language of the user's latest message, even when the documents are in another language. For more consistency, detect the language in code and insert it into the prompt as a variable. Then test with twenty Portuguese questions whose answers come from English documents. How many replies stay in Portuguese before and after the change?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Explain how quality and token counts differ across languages
- Make the model reply in the user's language reliably, including mixed input
- Decide between native generation and translation pipelines by testing
- Handle locale formats, formality and right to left scripts
- Evaluate each language with native reviewed test sets
Lesson plan
- 1 How languages differ for models Understand quality and token differences between languages. Start
- 2 Replying in the right language Keep responses in the user's language, even with English context. Start
- 3 Native generation or translation Choose an approach per language based on measured results. Start
- 4 Locale and interface details Handle formats, formality and scripts correctly. Start
- 5 Retrieval across languages Search documents in one language with queries in another. Start
- 6 Evaluating per language Build and review test sets for each supported language. Start
Try asking
About this tutor
For developers building LLM features for users who write in different languages, or mix languages in one message. You learn how quality varies between languages, why some scripts use more tokens (affecting cost, latency and context), how to make the model reply in the user's language reliably, when to translate versus generate natively, how to handle locale details such as dates, numbers and names, and how retrieval works across languages with multilingual embeddings. Most of all, you learn to evaluate per language with native speaker review instead of assuming English results carry over.
Reviews
Students can review a tutor after a paid lesson. Nobody has yet.
About the teacher
Structured output, tool calling and safe input handling for LLM applications that must behave predictably
9 tutors 308 lessons taught Sample
I teach the parts of LLM apps where free text has to meet real software: JSON that must parse, tools the model calls, images and documents coming in, and users who send things you did not plan for. I spent years writing integrations between messy systems, which taught me to treat every input as untrusted and every output as something...
See Greta's profile and tutorsMore like this
Other tutors on the same or nearby topics.