Skip to content
Anton GoncharovAI systems developer

Next.js SaaS website · sandbox, dashboard, API

SaaS: AI replies to WB and Ozon reviews

A marketplace seller needs reviews twice over: the reply affects the rating, and complaints show what to fix in the product. The service handles the routine — five-star reviews are answered automatically — but doesn’t hand the machine what needs a human: defects, breakdowns and accusations arrive as a task with a draft. Before publication every reply goes through the checks marketplaces really apply: no phone numbers, links or other platforms.

Scenario: Otklik service from Novosibirsk and its client, the Severny Dom home goods brand with stores on WB and Ozon — fictional: six products, twelve reviews in the queue and a stream of new ones.

Demo screenshot: SaaS: AI replies to WB and Ozon reviews

Open the demo

In the sandbox on the first screen, pick a review — the reply types out with analysis and checks. The seller dashboard and API docs are linked in the header; a trial, store connection and invoice appear in the owner dashboard.

Leads from the demo website are visible only to you, in the “Owner dashboard”: the server validates the form and stores nothing.

What the demo shows

  • Sandbox: an example or your own review, the marketplace, product, rating, name and brand tone — the reply types out in a card as on the marketplace

  • Review analysis: sentiment, 14 topics from size to delivery, complaints, accusations and caps lock; the reply addresses every complaint with product advice

  • Eight checks before publication: length, contacts and links, other platforms, stop phrases, answering the point, addressing by name, tone, similarity to past replies

  • Rule-based autopilot: what to publish immediately, what to send for approval, what becomes a task; defects and breakdowns always go to a human

  • Seller dashboard: a WB and Ozon queue with filters, “Run autopilot”, a live stream of new reviews, editing, another variant and publishing

  • A rule editor with a preview on a real review, brand tone, signature and your own stop phrases

  • Analytics: topics and complaints by product calculated from the queue, rating by month, the autopilot’s share

  • Plans matched by volume and number of stores, a payback calculator, an invoice by tax ID with check digit validation

  • API documentation: methods, parameters, examples in cURL, Python and JavaScript, webhooks, error codes and a console with the same checks and rate limit

  • An owner dashboard with trial requests, store connections and invoices

How to try it

  1. Reply to a review

    In the sandbox, choose “The lid leaks” or “Crooks” — see how the reply, tone and autopilot decision change.

  2. Break the checks

    Press “Edit”, type in a phone number or “we’ll refund you” and try to publish — the checks stop the reply.

  3. Run the autopilot

    In the dashboard, press “Run autopilot”, then “Live stream” — new reviews arrive and are processed on their own.

  4. Change the rules

    In the “Autopilot” tab, send five-star reviews for approval or add a stop phrase — the preview shows what changes.

  5. Send an API request

    On the API page, choose the “Phone in reply → 422” example and press “Send” — the console returns a check error.

No paid neural networks

There are no paid neural network requests: the reply is assembled by rules and templates in the browser — topic analysis by dictionary, phrases for the brand tone and complaints, advice about the specific product. In a production version a language model takes this place, while the analysis, checks and autopilot stay the same. The API console runs requests on a local implementation; store keys are checked only for format and are not sent anywhere. The service, brand and reviews are fictional.

The demo runs without OpenAI, YandexGPT or GigaChat API keys. Wherever a production system calls a model, there’s a local stub with the same interface — so the demo is free, always available and never makes things up.

What a production project adds

  • Reviews and reply publishing via the Wildberries seller API (Feedbacks API) and the Ozon Seller API, polled every 5 minutes with a queue and retries
  • Replies by a language model (YandexGPT or GigaChat) with the product card, brand rules and examples of good replies in the prompt; checks run after the model, as in the demo
  • Tasks for negative reviews go to a Telegram bot or the responsible person’s CRM, with a draft and a link to the review
  • Billing: plans and extra replies, invoices and acts via e-document exchange, card payment via acquiring
  • A public API with per-store keys, rate limits, a log and signed webhooks

Stack

  • Next.js 16
  • React 19
  • TypeScript
  • Tailwind CSS 4
  • Server Actions

Developer’s comment

Anton Goncharov

AI systems developer · Python

For business

  • Five- and four-star reviews get answered automatically within half an hour — the seller stops replying in the evenings, and buyers see a lively store.
  • Negative reviews don’t drown in the general stream: defects and accusations go to the responsible person as a task with a draft and aren’t published by the machine.
  • Review topics show what to fix in the listing: a size chart or daylight photos remove a whole class of three-star reviews.

For developers

  • Parsing, generation, checks and autopilot are pure functions with tests; the same code runs in the sandbox, the seller account and the API console.
  • Checks run after generation and don’t depend on the model: replacing templates with an LLM doesn’t weaken protection against phones, links and promises.
  • The account keeps the queue in localStorage via useSyncExternalStore — state survives a reload and syncs between tabs.

Tips

  • Tip 01

    Don’t auto-publish replies about defects and breakage, even on a five-star review: one unfortunate phrase about returns costs more than an hour of a manager’s time.

  • Tip 02

    Compare a new reply with previous ones: buyers notice identical replies under a hundred reviews before the marketplace does.

Want the same for your business?

Tell me about your task — I’ll tell you what from the demo fits, how long it will take and what I’ll need from you.

Other demos in this section

All demos
  • Next.js website · service landing page

    live

    Dental clinic landing page

    A one-page clinic website that brings in patients: a quiz suggests a treatment, a calculator prices the plan with installments, online booking shows doctors’ free time, and an AI consultant answers from the price list. The headline is A/B-tested; leads and goals are visible in the owner dashboard.

    • Next.js
    • React
    • TypeScript
  • Next.js website · B2B, multi-page

    live

    Manufacturing company website

    A multi-page website for a steel structures plant that sells on its own: a calculator prices a building with a live drawing, a builder assembles a technical specification, case studies filter by industry, you can apply for a job with a CV, and site search understands queries in plain words.

    • Next.js
    • React
    • TypeScript
  • Next.js website · catalog, map, comparison

    live

    Real estate catalog with smart search

    A real estate agency’s apartment catalog: type “two-bedroom near the metro under 12 million” and the search sets the filters itself, shows apartments on a city map, compares up to four and sends new ones to Telegram.

    • Next.js
    • React
    • TypeScript
  • Next.js website · AI translation, per-language SEO

    live

    An export website in four languages

    A website for an Altai honey exporter in Russian, English, Chinese and Arabic: translated page URLs, per-language SEO, a wholesale price calculator with delivery and a translation studio that checks the neural network’s draft against a glossary.

    • Next.js
    • React
    • TypeScript
  • Next.js website · customer account, AI assistant, Telegram

    live

    Customer account portal

    A print shop’s customer account: orders with production stages, layout approval with comments right on the layout, invoices paid via SBP, documents with an electronic signature, a new order with a price calculation, a chat with an AI assistant and a manager, notifications in Telegram.

    • Next.js
    • React
    • TypeScript
  • Next.js website · Python in the browser, AI mentor

    live

    Online school with a student account

    An online analytics school: a landing page with a Python task checked right in the browser, a level test with a personal curriculum, installments with a tax deduction estimate, a student account with homework and AI review, an AI mentor that answers from course materials and a certificate verified by QR code.

    • Next.js
    • React
    • TypeScript
  • Next.js website · booking by plate number, photo report, approval

    live

    Car service with booking and repair status

    A car service website: booking by plate number or VIN into free bay slots, a “what’s knocking?” diagnosis with likely causes and a price range, repair status with a photo report, approval of extra work and payment by link, and a garage with service history and a forecast of the next maintenance.

    • Next.js
    • React
    • TypeScript
  • Next.js website · case analysis, claim letters, calculators

    live

    Law firm with AI case analysis

    A law firm website: describe your situation in your own words right in the hero, get an AI analysis with an assessment of prospects, a claim letter in Word, court fee, penalty and limitation period calculators, a document checklist, case progress in a client account, a price builder and booking with a lawyer.

    • Next.js
    • React
    • TypeScript
  • Next.js restaurant website · menu, table booking, delivery

    live

    Restaurant with delivery

    A restaurant website with delivery: a menu with allergen and diet filters, AI dinner selection from a plain phrase, table booking on a floor plan showing free tables for the chosen time, a cart with delivery zones, a minimum order and promo codes, and order tracking with the courier on a map.

    • Next.js
    • React
    • TypeScript
  • Next.js fitness studio website · booking, memberships, account

    live

    Fitness studio

    A small-group fitness studio website: a weekly schedule with spots and two-click booking, a waiting list that moves you into the group automatically, memberships you can freeze from your account, an AI 4-week training plan from five questions and a personal account with attendance stats.

    • Next.js
    • React
    • TypeScript
  • Next.js IT conference website · program, tickets, networking

    live

    Conference website

    A two-day conference website: a program by room and track with a personal schedule, AI selection of talks without overlaps by role and free text, tickets in price waves with promo codes and a team discount, a QR code for every attendee checked at the entrance, and networking with matching and meetings during breaks.

    • Next.js
    • React
    • TypeScript
  • Next.js corporate logistics website · calculator, tracking, AI request

    live

    Transport company website

    A transport company’s corporate website: a calculator that compares groupage and a dedicated truck at once, shipment tracking with the truck on a map, AI parsing of a request from an email or waybill with missing fields highlighted, a map of warehouses and terminals with a storage calculation, a carrier application with an eligibility check, and company details with checksum validation.

    • Next.js
    • React
    • TypeScript