A project by Roman Larionovskii
Analytics and automation service for Ozon sellers
Sync with the Ozon Seller API, automatic financial reports with taxes, a Telegram bot and mini app.
No sign-up needed: the demo opens on a store with made-up data. The interface is in Russian.Web dashboard screens on a demo store with made-up data. Click to open full size. The interface is in Russian.
FBS and FBO orders and returns sync every 5 minutes, changes only. Open orders are rechecked twice an hour and the catalogue updates hourly. If Ozon hits its rate limit, the job is queued and retried. API keys are stored encrypted with AES-256-GCM.
Accruals, EAEU sales, compensations and reversals, settlements with tax at 6% or 8% of revenue or 15% of profit, depending on the regime. Reports build themselves 12 hours after the period ends. XLSX files are generated on the server, so the website and the bots serve the same file.
Messaging goes through a shared transport layer, so a new messenger plugs in without touching reports. A chat is linked with a one-time code, each report type has its own delivery schedule, and a send log prevents duplicates. The mini app opens via an HMAC signature of the launch data. It has an overview, stock, sending reports and labels to the chat, and personal notifications.
Nightly FBO stock snapshots and the change over the week. The supplier order is calculated from sales velocity, split between FBO and our own warehouse, and exported in the supplier’s format.
PDF labels with a barcode and the EAC mark, certificates of conformity and bulk filling via Excel.
A question from Ozon arrives in the chosen chats. A normal reply in the chat is published on Ozon. Everyone sees who answered, and a second answer is blocked.
A new bot user sees the product on a “live” demo store in one or two steps. The data is generated deterministically and refreshes every hour, and writes to the demo are blocked on the server.
Progress in story points: done out of planned. Data from Jira.
Epic start and due dates from Jira. One-day epics are shown as a dot.
Loading…
Task titles are shown as they are in Jira, in Russian.
Loading…
Every epic has a PRD: context, goals and non-goals, personas, FR and NFR requirements with MoSCoW priorities, metrics and risks. Stories follow the “As a… I want… so that…” format with Given / When / Then criteria. The backlog, sprints and timeline live in Jira, PRDs in Confluence. When an epic is cancelled, Jira automation closes its child tasks with the “Won’t do” resolution.
Someone who found the bot on their own got the reply “open the settings on the website”, and the mini app showed an error. The demo data was static, from January 2025.
FR-1Greeting on /startFR-3The /demo command sends a sample XLSX reportFR-6Sign-in as the demo user with a valid signatureFR-14Hourly refresh with a stable historyNFR-1Writes blocked on the server (DemoGuard)NFR-2No more than one file per 10 seconds per chatNFR-3HMAC-SHA256 signature; a forged one gets 4018 stories, 28 SP
The Ozon Seller API has no methods to approve or reject FBS return requests; they exist only for rFBS. The request text and photos are not available through the API.
The epic’s main value, handling requests right from the chat, can’t be built. Notifications without actions add little.
The epic was cancelled with the “Won’t do” resolution. Worth revisiting if Ozon adds the API.
Task status, story points, dates and the sprint come from Jira, the counters from the repository’s master branch. Task descriptions are never published.
A Node.js script builds data.json; the page reads it and redraws the live blocks once a minute.
Project authorRoman LarionovskiiProduct manager / e-commerce analyst. Experience, skills and contacts in the CV.