Hey LlamaIndex community! 👋 I'm an experienced dev looking to build a project to learn the framework in-depth.
I want to create an agent that helps create and refine business offers using natural language. Here's what I want it to do:
- Take an initial offer description and repeatedly prompt for missing info, if any (customer name, products, quantities)
- Validate inputs (e.g., no negative quantities, verify products/customers exist)
- Generate the offer in LaTeX format
- Allow questions about the offer
- Allow refinements (change or swap products, ...)
- Support offer finalization as an end state
I have some architecture/design questions I'd love input on:
- What's the best way to handle conditional LaTeX output vs Q&A responses? (Evaluation?)
- For offer creation/updates - should these be handled via response generation or function calls?
- How to properly manage state transitions (no offer → draft → finalized)?
- Should validation be its own LLM workflow step?
- Best practice for product/customer data - prompt injection, vector store, or filtered function calls?
Any guidance would be much appreciated! 🙏
