Self-Promotion
Show off what you're working on
Start new thread

I'm building an AI agent that investigates failed data pipelines. What would you need to trust it?

by•

Hi everyone, I'm Vijay, a solo founder building Arcyria.

When a data pipeline fails or a metric looks wrong, the investigation usually means jumping between logs, schemas, run history and a SQL editor. Arcyria puts that in one place.

What it does

- Connect sources and destinations, build SQL transformations, and monitor every run

- When something breaks, the Arcyria Agent looks at the pipeline, its runs, logs and schema

- It runs read-only SQL on the affected data and shows the query and the results

- It separates what it has confirmed from what it only suspects

- Any fix needs your explicit approval and respects your permissions

Where it is: a working MVP with no external users yet. I'm not claiming traction. I'm looking for the first teams to try it on real incidents.

Two questions for you

1. What was the last pipeline failure or wrong number you had to chase down, and what was the hardest thing to find?

2. What evidence would you need to see before trusting an AI's diagnosis?

If you're on a small data team and want to try it, reply here or email . Site:

3 views

Add a comment

Replies

Best

"It separates what it has confirmed from what it only suspects" is the line I'd trust. That's the same rule I hold my own agents to: nothing gets reported as working until the real thing has been checked, and anything that wasn't checked has to say "unverified". Showing the query next to the result does the same job.