← All tags
Tag
#llm
2 posts tagged with this topic.
Constraining a local LLM to valid JSON with GBNF grammars
Prompts ask a model to "return JSON". Grammars make it impossible to return anything else. How I used GBNF + constrained decoding to get reliable structured output on-device.
Designing a provably-offline app: privacy as a hard constraint
Lessons from building a local-first finance analyzer — how to make "offline" a guarantee you can prove, not a promise, and the architecture that falls out of it.