Scalabs AI — Agentic Content Platform
Role: Backend & infrastructure lead @ Biralo Studio
Stack: Dart, PocketBase, RabbitMQ, Chromium, Flutter
Multi-agent system that researches, writes, reviews, and publishes long‑form content automatically. I built the core backend and orchestration.
Here is mostly what I did in this product:
-
Scheduler & Pipeline
Standalone Dart cron service + main HTTP/RabbitMQ backend. Each stage (scrape → LLM → GitHub → filter → social) has its own queue with retries.
-
Scraping Stack
This workflow has four‑stage adaptive ladder per domain: HTTP → headless browser → proxied solver → residential IP. The system learns which stage works for each site.
-
External Integrations
Apify webhooks for social platforms (Facebook, YouTube, Reddit, LinkedIn) with duplicate prevention. This also has integrated cloudinary, R2 databases and webhooks for notifications, discord messages and telegrams.
-
Output & Delivery
Scheduled prompts that email results — duplicate runs are ignored automatically, with retries and a fallback email gateway. Initially supported multiple LLM providers; later OpenRouter as primary.
-
Ops & Diagnostics
Real‑time per‑domain and per‑worker success/failure tracking, queue depth monitoring, and detailed failure logs. Infrastructure includes automated GitHub PRs for generated articles, image hosting, secrets management, error tracking, ops alerts, and auto‑TLS via Docker + Traefik.
-
Frontend (Flutter)
Feature‑based architecture, Cubit state management, a clean service layer, consistent theming, and stubbed screens. Later handed off to a dedicated frontend engineer.
-
What I Learned
Hardest part wasn’t Dart — it was reliable system design, queue boundaries, graceful failure, resumable workflows, and load spreading.