I’ve implemented Temporal activities that pull content for each supported platform/content type and then feed that into an LLM with the bounty requirements. It’s not very elegant but for an MVP it seems fine.
Now that tool calls and structured outputs are more baked into LLM APIs, I have some refactors planned, to make the Workflow more “agentic”, but I don’t expect the user experience to change all that much; the biggest improvement will be being able to dive deeper into the content, assess images, etc.
Now that tool calls and structured outputs are more baked into LLM APIs, I have some refactors planned, to make the Workflow more “agentic”, but I don’t expect the user experience to change all that much; the biggest improvement will be being able to dive deeper into the content, assess images, etc.