Product-Quality Complaint Intake & Triage

The problem this kills

A product-quality complaint is not the same as a support ticket - and treating it like one is how recalls get missed. Right now the people who actually own product quality are stitching this together by hand: complaints arrive by email, phone, and forwarded support threads; someone retypes them into a spreadsheet; the lot or serial number is missing half the time; and the one complaint that hinted at someone getting hurt sits in a queue next to "the box was dented."

When severity lives in people's heads and routing depends on who happened to read the email, two bad things happen. Slow: a genuinely serious, potentially reportable event takes days to reach quality and regulatory. Noisy: low-severity gripes get the same alarm treatment, so the alarms stop meaning anything. Neither is acceptable when you're the team accountable for safety and for proving traceability later.

What you'll build

A focused internal web app for capturing and triaging product-quality complaints - distinct from general customer support. A complaint comes in (typed in, imported from a CSV, or forwarded from support), always tied to a specific product and lot/serial. AI reads it and proposes a severity level and complaint type, and explicitly flags anything that smells like a potential safety or reportable event. Then a quality reviewer confirms or overrides that call. Only after the human approves does the tool act: serious complaints route to quality/regulatory immediately, the complainant gets an acknowledgment email via Resend, and everything - serious or not - lands in a clean, exportable complaint log for trending.

What's inside the Implementation Plan

• It interviews you first. Before writing a line of code, the plan has the AI agent interview you about your actual complaint process: your products and how you name lots/serials, where complaints come from today, your real severity and complaint-type definitions, your reportability triggers, your volumes, and your messy edge cases. It reflects a short tailored spec back to you and waits for your thumbs-up - so you get a tool shaped around your QMS, not a generic template.
• A complete, copy-paste build: data model, intake screens, AI classification, the reviewer queue, escalation routing, acknowledgment email, and CSV export.
• AI severity/type classification with a plain-English rationale - and a hard rule that the AI only ever proposes; a person decides.
• Built-in safety-event flagging that links a confirmed potential adverse event over to your reportability workflow.
• Duplicate guards that catch the same customer + product + lot inside a date window before it becomes two records.
• A QMS-format CSV export so the log drops straight into the system you already report from.
• A "No API yet?" fallback so the whole thing is buildable today from a spreadsheet, with no integration required.

The governance it includes (this is the point)

This is built for a regulated quality function, so the controls are not optional add-ons:

• Login so only your team can open complaints.
• Row-level security so each organization only ever sees its own complaints.
• A complete audit trail - who logged it, what the AI proposed, who confirmed or overrode the severity, who escalated it, and when each happened.
• A hard human-in-the-loop gate: the AI drafts the severity, type, and safety flag, but nothing escalates, acknowledges, or commits to your log-of-record until a quality reviewer approves.
• Duplicate guards so the same complaint can't be triaged and routed twice.

Who it's for

Quality, regulatory, and customer-quality teams who must capture and triage product complaints with traceability - and who need the serious ones to reach the right people now, not after the weekend. If you can describe your severity rules in plain English and you have a spreadsheet of past complaints, you can build this.

You've got this - paste the first prompt and let the agent interview you.