Built an “on-demand analytics” demo (NL → SQL → chart → insights). Looking for feedback.

I’m experimenting with an on-demand analytics workflow where you can ask a question in plain English and get:

• generated SQL
• a chart/table
• a short insight summary (what changed + why it matters)

Examples it supports right now:

• “total revenue by region”
• “monthly revenue trend”
• “revenue by product category”
• “deal count by sales rep”

Demo: https://reports-ai.utils.product-led-growth.com/
Video walkthrough: https://www.youtube.com/watch?v=SaHZDKbKh1I&t=238s

What I’m trying to validate

1. What’s the minimum set of trust/guardrails you’d need to rely on this? (SQL visibility, allowlists, metric definitions, query limits, etc.)
2. Would you use this more for internal reporting or embedded analytics inside your SaaS?
3. What’s the biggest failure mode you’d worry about?

If you’re open to it, reply with your data stack (Postgres / BigQuery / Snowflake) + one “messy” question you wish you could answer quickly, and I’ll share how I’d constrain it to be reliable.