Averröes. The system an entire ops team now runs on.

Data people actually want to open

Every screen works in layers. A tile answers the question in one glance, a chart lets you explore, and one click drills down to the record behind the number.

And it is meant to be fun: advisors as rated cards, monthly trophies, publication streaks in 3D, challenge seasons. Ops teams use what they enjoy.

One place for every ops workflow

Thirty-one screens, five business segments. The modules I'm proudest of:

Subscription billing

Monthly advisor fees, collection rate, unpaid follow-up and free months in one view, exported in two clicks for accounting.

Transactions

Every deal followed from declaration to billing to payment, with the checks that flag a gap before anyone has to look for it.

ACC sections

Each account manager's pool, rankings, monthly podiums and trophies.

Catalogue

Every live listing as a card, filterable by city, cluster, type and advisor.

Promo & continuing education

Onboarding cohorts, Zoom sessions where only approved guests get in, reminders and attendance, end to end.

Built to hold up when nobody is watching

Decisions I made, each with a reason:

Three databases, three trust levels

One writable app schema, a read-only copy of the warehouse reached through a self-healing tunnel, and the live warehouse, read-only.

Access enforced at exactly three points

Role defaults live in code, per-user overrides in the database, and the page, the API and the menu all ask the same question.

Nothing outside trusts the browser

Anything sent to another platform is rebuilt by the server from the typed fields only. A tampered form cannot change an account type or backdate a contract.

Every card loads on its own

One card was ready in 165 ms and still waited 1.5 s for its slowest neighbour. Now each one streams in as soon as its own query answers.

Refuse rather than lie

If the warehouse comes back empty in the middle of a sync, the app stops and says so instead of sending zeros.

Migrations only add before a deploy

Removals ship after, in a separate step. I learned that one by repairing production.

29 workflows that keep it fed

Averröes prepares; n8n executes. The app composes a clean request (create this account, push these scores) and hands it to a workflow that talks to the outside platforms.

Behind it runs the advisor lifecycle as a state machine of about 280 steps, from recruitment and payment to contract, accounts, e-learning, activation, retention and offboarding, with escalation deadlines across teams. Thinkific, Slack, Zoom and Google Sheets all sit in that layer.

Two years of cleaning before the first chart

The part AI cannot do for you: learning what the data actually means. I cleaned, reconciled and modeled it with dbt for two years so every number on screen can be defended.

Example: I audited the 2,063-line advisor scoring model and found three structural flaws. Missing data was scored as poor performance, single-track advisors were dragged down by an empty second track, and only 17 of 89 computed indicators reached the published table. Half the network sat within 0.06 points. I built a tuning layer that replays the formula exactly, then fixes it in the open.

Eight sources. One reading.

Consolidated, reconciled and kept current in one place: HubSpot, Zendesk, YakeeyPRO, Divalto, Yakeey Marketplace, Zoho Mail, WhatsApp, and the PostgreSQL warehouse with dbt.