FeedbackForge

Inspiration

Most customer complaints are too noisy for engineers to investigate one by one—but those weak signals can be the first sign of a real incident.

We’ve seen cases where customers were already complaining about bad outputs or broken workflows while dashboards still looked “normal.” By the time metrics caught up, the issue had already impacted important users.

The problem is that early signals are scattered, low-confidence, and too expensive to investigate manually—so they get ignored.

What it does

FeedbackForge prioritizes and investigates customer tickets based on commercial impact—then turns the most important complaints into incidents teams can act on safely.

Using WunderGraph, it connects support tickets, Jira, CRM, product usage, metrics, and deploy history into one investigation graph. This lets the agent access everything it needs to determine whether separate complaints are actually the same underlying issue.

The agent then identifies the likely cause—for example, a prompt change, deploy, schema issue, or feature flag update—and recommends a fix.

Using Guild, actions are handled safely:

low-risk actions (creating tickets, notifying owners) run automatically high-risk actions (rollbacks, changes affecting many users) require approval and are fully audited

Using InsForge, the agent applies backend fixes directly—updating configs, adding detectors, creating runbooks, and patching simple workflow logic so similar issues are caught and resolved faster next time.

How we built it

We used WunderGraph to connect support tickets, Jira, CRM, customer ARR, product usage, metrics, and deploy history into a single investigation graph.

That graph gives the agent access to all the data it needs to do a proper investigation—understanding who is affected, the commercial impact, how many similar complaints exist, and whether the issue links to recent system changes.

We used Guild to control what the agent can do next. Low-risk actions run automatically, while high-risk actions like rollbacks require approval and are fully audited.

We used InsForge to let the agent make backend fixes directly—updating configs, adding detectors, creating runbooks, and patching simple workflow logic so similar issues are handled faster next time.

Challenges we ran into

Time management of such a big project

Accomplishments that we're proud of

Learning lots of new tools

What we learned

What's next for FeedbackForge

Making the agent capable of doing deeper investigations. We could only test on certain types

Built With

• guild
• insforge
• ts
• wundergraph