FirstFix

The problem

NovaLink's support desk handles a high volume of repetitive, low-complexity issues that customers could resolve themselves in under five minutes if the answer reached them in plain language for their exact device. Most contacts are first-line self-fixable, but the existing help page and PDF manual force the customer to translate their symptom into technical vocabulary, then leave no trace if the guidance fails.

Before FirstFix

Customers waited 6-11 minutes to reach an agent for issues like "Issuer Inoperative" on a POS terminal. Agents repeated the same instructions roughly 1,200 hours each month. After-hours merchants had no coverage at all. The knowledge to fix most faults already existed inside NovaLink, but it was not reachable at the moment of need.

What I built

FirstFix turns a free-text complaint into a diagnosed, logged, and actionable support ticket without a human agent picking up the phone. The customer describes the fault, receives a ticket reference, severity badge, root cause, and up to three numbered steps, then clicks either "This solved it" or "Escalate to a human agent."

• Lovable frontend. A mobile-first, no-login Help Center generated and hosted on Lovable, built from a written specification in about 25 minutes.
• n8n backend. The entire back end is a visual workflow: webhook, normalisation, LLM diagnosis, validation, Supabase logging, and response.
• Structured output. The model returns category, error code, root cause, severity, confidence, three steps, and an escalation flag in strict JSON.
• Ticket ledger. Every diagnosis is stored in Supabase with the exact guidance shown, creating a measurable record of which instructions resolve which faults.
• Clean escalation. When the AI cannot solve it, the already-diagnosed ticket with full context is handed to a human agent.

The outcome

Built as a Week 9 AI app project, FirstFix demonstrates that first-line support automation is now cheap enough to deploy in under two hours. The intended outcome is fewer queued calls, shorter wait times, and a data record that improves both the knowledge base and the product over time.

Stack

Next improvements

• WhatsApp channel. Add a WhatsApp Business entry point so merchants can describe faults from the same device they use for transactions.
• Confidence routing. Automatically escalate when model confidence falls below a threshold, before the customer sees the steps.
• Analytics dashboard. Show resolution rate, escalation rate, and top fault categories by device type.