Internal product · WebFull-cycle development3 months2026
Support Harixx — an internal AI platform for handling customer emails.
We built an internal tool from scratch where AI reads the thread, pulls order context from the database, and drafts a reply — so the team handles more tickets in less time. Stack: Next.js 16, MongoDB, Tailwind, Google Gemini.
Результати · Слайд 1 з 2
Client
Harixx
Niche
Customer support · Internal product · Marketplace ops
Timeline
2026 · 3 months
Service
Full-cycle web development
01 — Challenge
What the client came in with.
The Harixx team was processing customer emails across multiple interfaces in parallel: inbox, marketplace consoles, order spreadsheets. Order context had to be looked up manually for every ticket — response time kept growing, SLA didn't.
Harixx came to us with a product brief: build a single tool where the support team handles customer correspondence from the first email to ticket close — without jumping between inbox, spreadsheet, and marketplace console.
The team was serving several businesses at once, each with its own ticket flow, reply templates, and SLA. Order metadata — ASIN, carrier, tracking number, delivery date — lived in different systems and had to be checked manually for every request.
The goal wasn't to build yet another ticketing system, but to ship a tool where AI takes the routine off the operator and leaves them in control of the final reply.
The starting requirements were concrete:
- 01Multi-business inbox. Several Harixx brands in a single workspace — with filters by business, manager, and status.
- 02AI that understands the thread. A Gemini assistant reads the whole conversation, classifies the request type, and drafts a reply with the order context.
- 03Order context inside the thread. ASIN, delivery status, tracking — pulled in next to the conversation, no jumping to external systems.
- 04SLA discipline. Tickets older than 3 hours auto-escalate; the dashboard shows average response and resolution time in real time.
02 — Approach
What we did differently.
We worked as a product team with weekly releases. The focus: get to a working version for the support team fast, then iterate on real data instead of assumptions.
Phase 1 — Architecture and foundation. Set up a Next.js 16 App Router monorepo, picked MongoDB Atlas as the primary store for the document-oriented thread model, and built a multi-tenant data schema (Business → Inbox → Thread → Message) with access isolation between businesses.
Phase 2 — Inbox and statuses. Built a four-status workflow (New / In Progress / Escalated / Resolved), real-time UI updates when new messages arrive, search by order number, filters by business and manager, and a status audit log.
Phase 3 — The Gemini AI layer.Integrated Google Gemini for three scenarios: request-type classification, reply drafting grounded in the full thread history, and an “Auto-reply candidate” flag for tickets that can be closed with an automated reply after the operator confirms.
Phase 4 — Order context and SLA dashboard.Spun up an integration layer with marketplace APIs: order metadata (ASIN, carrier, tracking number, delivery by) is pulled in next to the conversation in real time. Built a dashboard with metrics — total threads, avg resolution time, avg first response time, escalated >3H, autoreplies, drafts, currently open.
The product went into production use by the Harixx team 3 months after kickoff — no release slippage.
Виконані послуги
What we actually delivered.
Five tracks that together produced an internal product ready for daily use.
//01
Architecture and infrastructure
Next.js 16 App Router + MongoDB Atlas, server actions, a multi-tenant data model with access isolation, role-based authentication.
//02
Design system and interface
A custom component system on Tailwind CSS v4 in the Harixx cream/coral brand palette. Responsive UI tuned for working with lists of hundreds of threads at once.
//03
Gemini AI assistant
Request-type classification, reply drafting grounded in the full thread context, an “Auto-reply candidate” flag for bulk processing of routine requests.
//04
Order context inside the thread
Marketplace API integration: ASIN, carrier, tracking number, and delivery date pulled in real time next to the conversation. Zero tab-switching.
//05
Dashboard and SLA monitoring
Team metrics (volume, resolution time, first response time, escalated >3H, drafts, currently open), charts by date and business, automatic escalation past 3 hours.
Результати
Numbers from Support Harixx.
The product has been in daily use by the Harixx support team since spring 2026.
3 mo
From zero to production. The team switched to the new tool 3 months after development started.
0 → 1
A brand-new product. Replaced working in the inbox, spreadsheets, and marketplace consoles.
AI-native
Gemini in every thread. From request classification to drafted reply.
Product features at a glance
As of May 2026AreaWhat was shippedStack
Inbox workspaceMulti-business inbox with New / In Progress / Escalated / Resolved statuses, search by order number, filtersNext.js 16 + server actions
AI assistantRequest-type classification, reply draft grounded in the thread, auto-reply candidateGoogle Gemini API
Order contextASIN, carrier, tracking number, delivery date — inside the thread in real timeMarketplace API + MongoDB
SLA dashboardVolume, avg resolution / first response, escalated >3H, drafts, currently openCustom analytics on MongoDB
Auth & multi-tenantData isolation by business, roles and permissions, action audit logNext.js auth + MongoDB
Trusted by
Over 45+ companies already work with us
“
Every reply used to start with hunting for context across three systems. Now AI prepares the draft and the order data is on the same screen. Time per ticket has dropped noticeably, and the team finally sees SLA in a single dashboard.
HX
Harixx Support Team
Customer support · Harixx, 2026
Contact
Want the same result?
Send a request — we'll book a free 30-minute call, run an express audit, and show you what we can do with your product.
✓
Free consultationon the first call∅
No obligations— just a conversationSubmit a request
Takes a minute to fill out. Three fields and we're already thinking about your project.