Generate a AV event equipment list from user input without over-engineering inference?

Questions
1

Problem Description

I’m building a system that should generate a realistic AV / event-production equipment list when a user describes a new event (e.g. corporate conference, audience size, venue type).

What I already have

  • A real-world reference equipment list from an actual event (stage, rigging, lighting, cables, accessories, etc.).

  • The list is structured into blocks → items.

  • I experimented with adding AI-generated “reasoning”, triggers, confidence scores, etc. for each block and item.

My goal

The end goal is simple:

User enters details about a new event → system outputs a realistic equipment list that an AV company would actually use.

Where I’m stuck / confused

I went down a path where I tried to:

  • Infer why each block exists

  • Infer why each item exists

  • Create schemas for triggers, justifications, confidence

  • Chain multiple AI agents (event → block → item inference)

But this now feels over-engineered and brittle, especially since:

  • I only have one real reference event

  • I don’t have a database of past events

  • The “reasoning” feels artificial and not clearly useful for generation

I’m questioning whether:

  • I should even be doing inference at this stage

  • Or whether I should treat the reference list as a template / anchor and directly generate new lists by adapting it to user input (audience size, event type, venue)

What I’m asking the community

I’d love advice on:

  1. Architecture

    • Is it better to:

      • infer rules from the reference event, or

      • directly generate new equipment lists by adapting a reference example?

  2. Data strategy

    • If I don’t have historical data yet, how would you approach generation?

    • Is one high-quality reference enough to start?

  3. AI usage

    • Should this be:

      • one generation step, or

      • multiple reasoning/inference steps?

    • At what point does “explainability” actually become useful?

  4. Over-engineering check

    • What’s the minimum viable approach that still produces realistic results?

Constraints / Notes

  • This is not a recommendation system yet

  • I don’t need perfect explainability right now

  • Realism and usability matter more than fancy reasoning graphs

  • Quantities can be rough initially

2

It sounds like you’re overthinking it! Since you already have a solid reference list, try focusing on mapping user input (like “corporate conference”) to specific blocks and items in your existing list. You could use a simple JSON structure to represent the mappings, making it easy to generate the equipment list based on user input.