DEV Community

Rushikesh Bodakhe
Rushikesh Bodakhe

Posted on

What I Shipped Today: Evidence-Based AI Answers for Database Schemas

Today’s update is small on the surface, but important for trust.

I’ve been building Vizora, a schema-first platform that turns database schemas into diagrams, documentation, and understanding.

One recurring problem kept showing up during testing:

AI answers were often almost right — and “almost right” is the most dangerous kind of wrong.

So today, I shipped a fix for that.

The Problem I Wanted to Solve

Most AI assistants:

Give confident answers

Hide assumptions

Force engineers to double-check everything

That creates a debugging tax:

You ask a question

You get an answer

You still need to verify it manually

At that point, trust drops.

I didn’t want Vizora to become “just another AI chat”.

Today’s Update: Schema-Constrained, Evidence-Based AI

I updated the Ask Schema feature so that every AI answer is now verifiable.

New Rule (Non-Negotiable)

Every AI response must explicitly show:

✅ Schema version used

✅ Referenced tables

✅ Referenced columns

✅ Relationships involved

If the schema doesn’t contain enough information, the AI must say so clearly.

No guessing. No filler.

Example

Question

What breaks if I remove user_id from orders?

Answer

Removing orders.user_id breaks the relationship between orders and users, affecting ownership linkage.

Evidence shown

Schema v14

Tables: orders, users

Columns: orders.user_id, users.id

Relationship: orders.user_id → users.id

Now you can trust the answer at a glance.

Why This Matters

This update directly reduces:

AI hallucinations

Verification time

Context switching

“Let me check the schema myself” moments

The AI is no longer a guessing assistant — it’s a schema reasoning layer.

What I Deliberately Didn’t Add

❌ No generic chatbot

❌ No free-form AI answers

❌ No runtime behavior speculation

❌ No query-level debugging

Vizora stays focused on structure and understanding, not execution.

What’s Next

Next updates I’m working on:

Schema quality & risk insights

Auto onboarding guides for new developers

Better schema diff visualizations

Still keeping the scope tight and developer-first.

If you’re building tools with AI inside, I’m curious:

How do you handle trust?

Do you show evidence, or expect users to verify?

Happy to discuss in the comments.

Thanks for reading 👋

Top comments (0)