Enterprise
Learning
Platform

The Problem: The product had to be demonstrable end to end — real enrolments, approvals, scores, transcripts — long before the client's backend existed. Mock JSON would have meant building every screen twice, and throwing the data logic away when the real API arrived.

My Solution: SQLite compiled to WebAssembly, running in the tab. On boot it restores the last snapshot from IndexedDB, runs ordered, append-only migrations, and seeds if empty; every write schedules a debounced snapshot, with a pagehide flush as a safety net. Seventeen repositories are the only code allowed to write SQL, and screens depend on their interfaces through injection tokens — so moving to an HTTP backend is a change of provider bindings, not a rewrite.

The Problem: Enterprise customers always want one more field — and a different field for each role. Hard-coding forms means a code change, a schema migration and a deployment for every request.

My Solution: Forms and tables render from metadata. Shipped manifests are the defaults and the database is the truth: an administrator renames, hides, reorders and adds fields in Screen Studio, and sets per-role view and edit rules. New fields claim one of twenty reserved columns on each core table, then spill into a JSON column — no migration. Identity fields are locked so they can be renamed but never removed, which keeps the configuration from breaking the product.

The Problem: Completing a course touches everything: the completion date, a rolled-up score, skills on the learner's profile, an expiry from the renewal rule, an audit row, notifications. If three screens can each mark something complete, they will eventually disagree.

My Solution: One recompute() in the progress service is the only place a status becomes COMPLETED, and it does all of the above in one pass. Activities unlock through sequencing and prerequisite checks, assessments grade against a pass threshold, and ownership is enforced on the row itself — the player refuses an assignment that isn't yours, because a route guard alone isn't enough.

The Problem: The design called for frosted-glass tiles and a gliding tab pill. Live backdrop blur visibly pops whenever an ancestor is transformed or faded — which is exactly what a tab transition does — and switching panels caused the page to jump as heights changed.

My Solution: The glass is built from layered translucent fills, hairline borders, inner highlights and a specular sheen over a soft page wash — no live blur at all. All tab panels stay mounted in a single grid cell, so the container is always the tallest panel and never jumps; switching is an opacity crossfade with a small glide, while one pill slides to the active tab's measured box. Every colour, size and spacing value is a token, and layout uses logical properties throughout, so the interface is ready for right-to-left languages.