Athlete memory

One persistent profile that follows the athlete across coaches, clubs and seasons. The athlete owns it, the AI maintains it, and only the people they trust ever see it.

Most software treats each user as a clean slate. When a player switches coach, signs up to a new club, or asks an AI tool a question, they start over from scratch. Avelok refuses that. We treat the athlete as the centre of gravity, and we keep one continuous record of who they are as a player — across every coach, club, parent and AI conversation in their career.

We call that record the athlete memory.

What's actually inside

The memory is not a chat log. It's a structured, opinionated picture of the athlete:

Skill profile

Per-skill rating, last-trained date, recent trend, sample drills the AI can pull when asked.

Recent context

The last 5–10 sessions and matches in plain language, with what was tried and what changed.

Body & mind notes

Anything the athlete or their coach flagged: a sore wrist, a mental block on backhand opens, a sleep schedule shift.

Goals & plans

What the athlete is working towards this month, this season, and what's queued for the next session.

The AI builds drafts of all four. A human edits and approves them. Drafts that nobody confirms within roughly two weeks get archived rather than promoted — we'd rather lose a guess than carry a wrong one forward.

Who owns it

The athlete. Always.

This is not a slogan. It's a hard rule baked into how the data flows:

  • Every coach, parent or club only sees the slices the athlete (or their guardian, for under-13s) explicitly granted.
  • When an athlete leaves a coach, the coach's view of that athlete's memory goes read-only and then expires after 90 days unless renewed.
  • When the athlete deletes their account, the memory and all derived AI summaries go with it — including coach-side caches.

Why it's not just a chat history

A naive AI memory would be: "save every conversation, retrieve relevant chunks." That fails for sport for three reasons:

  1. Sport is rhythmic. What you did three months ago matters less than what you did last week, and what you did last week matters less than how you felt today. The memory is time-decay-aware out of the box.
  2. Sport has structure. A "loop against backspin" is a thing. So is "footwork on the third ball." We index against a sport-specific skill tree — currently table tennis, with badminton and swimming on the roadmap — instead of treating every word as equal text.
  3. Sport is multi-author. The coach writes, the player writes, the parent writes, the AI writes. Without explicit role-tagging the memory becomes a contradictory blob. We tag every piece of context with who said it and when, so the AI can resolve conflicts ("coach said X on Tuesday, player said Y on Wednesday") instead of averaging them.

The two-week rule

Anything the AI inferred but no human confirmed within ~14 days gets soft-archived. It can be restored, but it stops affecting new AI answers. This is the single biggest design decision we made to avoid memory drift — the kind where an AI ends up arguing with a player about something the AI itself made up two months ago.

How to read it as a coach

The athlete memory is not a wall of text. In Coach Studio you see one player at a time, with four collapsed cards (skill / context / body & mind / goals). Each card has an edit button — you can rewrite, delete, or promote anything the AI suggested. There's also a timeline view if you want chronological reading; most coaches use it once a month, not daily.

How to read it as a player

In the Mini Program, the memory is mostly invisible by design. You'll see a small Profile tab with two things: your current strengths and your current weaknesses, in your own words (or your coach's, if they wrote them). You can tap either to expand. Most players never need to scroll deeper — the memory is there to make your AI answers and your coach's plans better, not to be read.

What's next