Live skill-tracking system behind a game-like experience score

A live tracking system that reads how a person is doing across five work skills (independence, quality of output, communication, speed, and an overall combined score), and keeps that read updated as they keep working, instead of only grading them once at the end. That read is turned into experience points, a running score, like a video game's XP, that goes up as the person's actual skill goes up, so a person's number reflects their real skill level, not just how many things they've clicked through.

Most learning systems grade individual assignments but never build a running picture of the person doing them. To be useful for something like matching someone to a real job opportunity, a system needs a live read of the qualities that actually matter to an employer: how independently someone pushes through the unknown, how well their work holds up, how clearly they communicate, how fast they move. That read has to keep updating as they work, not sit frozen from a single final test. Without it, an experience-point score is just a vanity number that goes up the more you click. With it, that same number becomes a real signal of where someone currently stands.

Every meaningful interaction (a chat reply, a note submitted, a voice session, a piece of work turned in, a full task completed, a presentation) sends its raw signal into this system. Two paths process it in sequence. A cheap path is deterministic and runs on every single interaction: it reads the person's current skill snapshot, picks a base number of points for that kind of interaction, multiplies it by a tier tied to their current overall skill level (the multiplier ranges from 1x for someone struggling up to 10x for someone excelling, so the points someone earns actually reflect their level, not just their activity), and adds it to a running total. A heavier path runs less often, whenever a bigger piece of work finishes (a full task, a presentation, an onboarding conversation). It runs an AI review that reads the full transcript, the grading criteria, the person's prior history, and the current skill snapshot, then rewrites the skill snapshot with a fresh verdict per skill. The very next interaction after that already earns points based on the freshly updated snapshot, so the score always reflects the latest read of the person, not a stale one. This same tracking system is the shared foundation three other projects on this page build on: the mood detector, the manager-relationship tracker, and the per-person running narrative.

Tech lead. Owned the architectural design of the system, the cheap deterministic path that runs on every micro-event, the heavier AI-review path that fires when a bigger piece of work completes, and the scoring rule that ties the point multiplier to actual skill level.