Live: scrimly.xyz · Stack: Django · React · PostgreSQL · Redis · Celery · Docker · Supabase Auth
A full-stack SaaS platform sole-developed over ~4 months. Scrimly handles the full lifecycle of a mobile esports tournament — from org setup and registration to live check-in, scoring, and public leaderboards.
No existing platform owned this workflow for battle royale games (COD Mobile, PUBG Mobile, etc.) — scoring was done manually in spreadsheets. Scrimly automates it end-to-end, including AI-powered score extraction from in-game screenshots.
- Multi-tenant architecture — orgs with Owner → Admin → Manager → Member RBAC, resource isolation, and per-plan limits enforced at the API layer
- Registration engine — draft/submit/approve workflow with waitlists, slot management, Discord OAuth verification, blacklisting, and withdrawals
- Billing system — Paddle integration with tiered subscriptions, free trials, credit-based AI usage, webhook processing, and a full billing portal
- AI score extraction — Vision AI (Qwen VL) parses game screenshots; extracts all team and/or player metrics in 1-2 minutes instead of manual entry over hours
- Configurable scoring — per-tournament scoring rules, configurable metric multipliers, player-level metrics, manual adjustments, and CSV export
- Production ops — self-hosted VPS with Traefik reverse proxy, zero-downtime deploys, hourly Cloudflare R2 backups, and Sentry error monitoring
- Test coverage — 1,783 backend tests (Pytest) + 126 client tests (Vitest) covering registration flows, scoring logic, billing, and permissions
The organizer dashboard surfaces active registrations, recent notifications, and pending actions across all their tournaments.
Each tournament has a dedicated admin hub with stats, schedule, and configuration at a glance.
Organizers manage registrations across status tabs (Pending, Registered, Waitlist, Rejected, No Show) with bulk approval, copy-to list, and per-row actions.
Teams see their full registration state — approval timeline, manager info, emergency contact, and full player lineup with IGN, Discord handle, Twitter, and preferred role.
Organizers configure tournaments through a 6-step wizard. Title, slug, game type, region, logo, and banner are set in step 1.
Step 4 lets organizers define exactly what information players and teams must provide at registration. Fields are configurable per-tournament, with support for custom fields.
Account requirements, Discord server membership, team allowlists, granular auto-approval toggles, and mid-tournament change controls — all configurable per-tournament.
Tournament directors configure multiple leaderboards per tournament — team standings, player stats — with visibility toggles and sorting rules.
The public-facing tournament page shows tournament info, schedule, registration requirements, and embedded leaderboard with full standings.
Tournament directors can preview individual leaderboards from the admin panel. The preview modal shows full standings with Refresh and Download options.
The score entry modal shows all registered teams in a 3-column grid. Each team can be expanded to enter individual player metrics alongside team-level metrics.
The standout feature: organizers upload match screenshots and Scrimly's AI (Qwen VL) extracts all team metrics automatically.
Step 1 — Upload game screenshots (up to 8 files)
Step 2 — AI processes screenshots
Step 3 — Review extracted results before applying (25/25 teams found)
| Layer | Technologies |
|---|---|
| Backend | Python 3.13, Django 5.2, Django REST Framework |
| Frontend | React 19, TypeScript, TanStack Router/Query, Tailwind CSS, shadcn/ui |
| Database | PostgreSQL (soft deletes, row-level constraints, optimized indexes) |
| Cache / Queue | Redis + Celery (async tasks, AI job processing, rate limiting) |
| Auth | Supabase JWT with JWKS validation (Discord OAuth for player verification) |
| Billing | Paddle (subscriptions, webhooks, credit system) |
| AI | Qwen VL via Alibaba Cloud (vision model for screenshot parsing) |
| Infra | Docker Compose, Traefik, self-hosted VPS, Cloudflare R2, Sentry |
| Testing | Pytest (1,783 tests), Vitest (126 tests), factory_boy, coverage reporting |
Why self-hosted over managed cloud? To demonstrate full-stack ownership — Traefik config, SSL, backup automation, container orchestration, zero-downtime deploy strategy.
Why Paddle over Stripe? Paddle acts as Merchant of Record, handling global VAT/tax compliance automatically. Better fit for a SaaS targeting international orgs.
Why Qwen VL for AI extraction? Higher accuracy on game UI screenshots (HUD layouts, non-standard fonts) than alternatives tested. API cost structure also maps well to a credit-based billing model.
Why DRF over alternatives? The registration and scoring logic involves deeply nested validation, state machines, and permission checks at multiple layers. DRF's ViewSet + serializer pattern and mature ecosystem kept complexity manageable across 1,700+ tests.
Source code is private. Staging demo available — contact me for access.