I'm building an eval system where PMs and other non-technical folks need to iterate on prompts through a web interface. I really don't want to rebuild all the nice things the BAML VSCode extension provides - syntax highlighting, rendering prompts with actual data, etc. Question: What can I reuse from BAML to avoid reinventing the wheel? Are there any headless components, APIs, or patterns I can leverage? Also curious - has anyone tackled something similar? Not just BAML-specific, but any approaches/tools for web-based prompt editing that play nice with existing LLM tooling?

Take a look at this - For that type of tool BAML certainly is a solid contender for your inference layer but I think you would end up finding you are still lacking tons in the workflow, GUI layers and would end up rolling your own thing anyways: https://mirascope.com/blog/prompt-management-system#1-lilypad "Prompt Management" is this pillar in the LLMOps domain.. (Agenta, PromptHub, Langfuse) .. some "Eval/Trace" tools will also inlcude things like this - but there are so many features interwoven ** some that make sense.. but the inference of the prompt is one of the smallest/simpler pieces.

we have also been considering making a more available react component that virtually one can use to render / play around with a BAML codebase. Is that something that would be helpful? (That is the same component that powers VSCode + https://www.promptfiddle.com )

yes, that would be very helpful

awesome we'll note this and share with the team!