OwnURstyles

Guide

The 35 AI jobs of a community fashion brand

Most “AI workforce” maps are marketing pages with a waitlist. This one is different: every job below is running code in the OwnURstyles backend — 7 departments, 35 jobs, one shared database as the company brain.

CompanyBrainDesignPipelineCommunity& VotingCommerceMoney& PayoutsIntelligenceSupportMarketing& Ops
01

Design Pipeline

From raw upload to printable artwork.

Upload intake

Presigned PUT URLs straight to object storage, SHA256 dedupe so the same file never uploads twice.

Vision analysis

A vision LLM reads every upload once and caches palette, detail level, and recommended shirt colors — the chatbot reuses it forever.

Background removal

A Python rembg sidecar strips backgrounds in-process; opaque uploads skip it automatically.

Mockup rendering

OpenCV pipeline: paste into the print box, displace along fabric folds, blend with the shirt's lighting map. Pixel-exact, no diffusion guesswork.

Chat refinement

The designer talks to an AI assistant ('make it black, bigger, left chest') and every reply becomes a new version with real print parameters.

Print-ready generation

When a design is approved, a queue job renders the production file before any money moves — unprintable art never reaches a drop.

02

Community & Voting

Demand validation before a single shirt is printed.

Weighted scoring

Vote = 1, wishlist = 3, share = 5. A queue worker recomputes scores, never the request path.

Auto-approval

Cross the threshold and the design promotes itself to production — no human gate.

Comments & moderation

One-level threads with profanity filtering and soft deletes.

Follows

Fans follow designers; follower counts feed the leaderboard.

Vote integrity

Partial unique indexes stop the same user, email, or IP-day from voting twice.

03

Commerce

Pre-orders, payments, and fulfillment signals.

Drop creation

Approved designs get a Drop with run size, price, and pre-order minimum — one per design, enforced by the database.

Pre-order funding

The drop flips to FUNDED only when enough pre-orders land. No demand, no print bill.

Checkout

Razorpay orders with price and royalty rate snapshotted at purchase time — history never rewrites.

Webhook settlement

HMAC-verified payment webhooks, idempotent by provider event id, mark orders paid exactly once.

Order status

Every state transition is explicit; the support chatbot reads the same rows the customer sees.

04

Money & Payouts

Designer earnings, calculated and paid automatically.

Royalty ledger

Append-only ledger entries on every paid order: 10% / 12% / 15% by lifetime units tier.

Tier computation

Royalty tiers derive from lifetime units sold — never stored, never stale.

Auto-payouts

A nightly sweep finds designers over ₹2,000 and fires RazorpayX payouts with idempotency keys.

KYC gating

Payouts hold until KYC is verified; admins review submissions in the console.

Payout polling

A 30-minute cron backstops provider webhooks so no payout is stuck in limbo.

05

Intelligence

The company brain — live answers from live data.

Ask the brain

Admins ask plain-English questions; the AI writes a read-only SQL query, a guard rails it, and every query lands in an audit log.

Nightly trend report

A cron aggregates 24h of votes, wishlists, and shares, then an LLM writes the insights and emails every admin.

Leaderboards

Daily, monthly, and all-time snapshots; top-3 designers earn cash bonuses.

Ad analytics

UTM-tagged clicks tie marketing spend to designs that actually convert.

Audit trail

Privileged AI actions are append-only logged — who asked what, which SQL ran, how many rows came back.

06

Support

Answers from your real account data, not a FAQ page.

AI support chat

The assistant calls read-only tools scoped to the caller — it can see your orders and royalties, never anyone else's.

Order lookup

'Where is my order?' reads the actual order row and answers with its real status.

Royalty explainer

Designers ask what they've earned and get the ledger total plus how the payout threshold works.

Human escalation

Refunds, disputes, and bans route to a human — the AI knows what it shouldn't touch.

07

Marketing & Ops

The brand talks while the team sleeps.

Auto social posts

The moment a design is approved, a worker writes the Instagram caption and queues the post — no social manager required.

Drop announcements

Broadcast emails to opted-in users when designs they validated go on sale.

Vote reminders

Weekly cron nudges the community back to open ballots.

New-user alerts

Admins get a notification on every sign-up.

Unsubscribe hygiene

Per-category opt-outs with a suppression list, enforced before every send.

See it working

Upload a design, vote on one, or ask the support assistant about an order — every feature on this page is live in the product.