Three-line summary - FINNS shows you one sentence instead of more numbers at the end of a workout. Something like "Today leaned recovery — Z2 made up 64% of the session." - That sentence is generated from 12 swimmer personas (endurance type, sprinter, technician, all-rounder, and so on) combined with your accumulated history. - It is not an LLM chatbot. The coach runs on a rule-and-statistics model plus a natural-language template — fully on-device, with no personal data leaving the phone.
1. Why a "Coach Note" Instead of Just Numbers
When a swim session ends, your watch hands you a stack of numbers:
- Distance: 1,200 m
- Time: 38 min
- Average HR: 142 bpm
- SWOLF: 38
Looking at those four lines, can you answer "Was that a good session or not?" instantly? Most people can't. Averages hide intensity shifts. SWOLF varies by stroke, but you only see one figure. And critically, the comparison to last time is missing — there's nothing to anchor "today" against "your usual."
That missing context is exactly what most swimmers end up reconstructing in their heads on the locker-room bench, often inaccurately.
FINNS's coach note folds that context into a single line:
- "Today was a recovery session. Z2 share was 64% — higher than your usual 45% — and freestyle SWOLF held steady. A good rhythm for bouncing back."
- "Looks like an interval set. Z5 was hit three times, and your between-rep HR recovery was 12% faster than usual. Stroke economy held up well."
- "Reads like a lesson main set. Breaststroke SWOLF ran 4 points above your baseline. Worth checking your glide length."
Each of those notes is short by design. The goal isn't to write an essay — it's to give you a single sentence you can read in the locker room while you towel off, and walk away knowing exactly what kind of session today was.
2. The 12 Swimmer Personas — Find Your Type
The tone of the coach note is different for each person — because the same data should be read differently depending on what kind of swimmer you actually are.
12 personas (three axes × four quadrants)
- Distance vs. Intensity axis: Endurance ↔ Sprint
- Technique vs. Conditioning axis: Technique ↔ Conditioning
- Routine vs. Variety axis: Structured ↔ Free
Combining the three axes yields 12 persona slots. A few examples:
- Endurance type — long distances at a consistent pace
- Sprinter — short and hard, with frequent intervals
- Technician — consistently low SWOLF across multiple strokes
- All-rounder — every stroke covered evenly
- Lesson-driven — high attendance during class hours, structured menus
- Early-morning regular — high share of pre-5 AM sessions
- Weekend-focused — light during the week, long on weekends
- Masters challenger — frequent personal-best updates
… and four more for a total of 12.
How a persona gets assigned
- The model looks at your most recent 12 weeks of workouts and assigns the persona automatically (not user-selected).
- The 12-week window is rolling, so as new sessions come in, older ones fall off. This makes the persona reflect your current training reality, not what you were doing a year ago.
- If your pattern shifts, your persona shifts with it — "In the last 4 weeks, your tendencies have leaned more sprinter."
- You can also lock a persona manually. (Example: "I want to focus on distance" → pin Endurance type.)
- A locked persona doesn't stop the analysis — it just keeps the coach note framing consistent while you train toward a goal.
Why 12, not 8?
An earlier version of the system had 8 patterns, but real swimmers don't fall cleanly into 8 boxes. A masters athlete who only trains at 5 AM behaves very differently from an evening sprinter, even though their HR distributions look similar. Adding the Routine vs. Variety axis captured those time-of-day and scheduling patterns that the original 8 missed.
3. How the Coach Note Is Generated
3.1 Not an LLM — rules plus statistics
To head off a common misunderstanding: FINNS's coach note does not call ChatGPT or any other external LLM API.
- The session's quantitative metrics (zone distribution, stroke share, SWOLF mean/std, pace variability, etc.) are extracted.
- Those values are compared against your own accumulated baseline to surface deviation patterns.
- A pre-designed sentence template in Korean, English, Japanese, and several other languages is filled in with the right variables.
- Only the tone and the emphasis change based on your persona.
The diagnosis spans eight areas — pace, heart rate, efficiency, stroke, frequency, volume, lesson patterns, and individual traits. How the coach turns those deviations into concrete Try-This drills (complete with distance, sets, and rest intervals) is covered in detail in Part 11.
3.2 What that buys you
- It's fast — the note appears the moment the workout ends. No API round-trip.
- Personal data stays on the device — every calculation runs locally.
- It's reproducible — the same data produces the same note. No drift between runs.
- No ad dependence — there's no LLM bill to offset, so the app doesn't need ads to survive.
3.3 What it can't do
- It doesn't chat. It is not a chatbot. One note per workout, one line.
- It is not creative. Templates are templates — tones can repeat occasionally.
- It doesn't take free-form questions. It cannot answer "Why is my breaststroke slow today?" in a conversation.
- It doesn't replace a human coach. Form feedback, drill prescription, and technical correction still need a coach watching you swim.
This limitation is a deliberate trade-off. Keeping personal data off external servers was the priority. An LLM-based coach would mean every workout summary makes a network call, and every network call is a chance for data to leak — by design, by mistake, or by policy change in someone else's terms of service. We chose the slower-evolving but fully local path.
4. Weekly, Monthly, and Yearly Reports
Separate reports are generated automatically every Sunday, every month-end, and every year-end.
Weekly report — example
- "Three sessions / 3,400 m total / 78% freestyle / Z3 reached 4 times this week."
- "Vs. last week: distance +15%, average SWOLF +1 (a slightly heavier week)."
- "Suggestion: add one recovery session next week."
Monthly report — example
- Cumulative distance, average pace, and SWOLF trend per stroke
- Lesson vs. free-swim distribution
- Newly earned medals
- AI coach paragraph: a one-paragraph summary tied to your persona
Yearly report — example
- Annual distance / distribution by stroke
- Monthly workout-frequency heatmap
- PB update timeline
- "Sentence of the Year" — a persona-aware annual summary
5. Real-World User Scenarios
Scenario A — Office worker, lesson-based
- Tuesday/Thursday evening lessons, Saturday free swim
- After 4 weeks: "Lesson-driven type — adapting well to a structured menu. Adding more Z3+ time on Saturday free swims would compound nicely."
- Medal track: lesson-attendance bronze, on the way to silver
Scenario B — Early-morning masters
- Weekdays 5:30 AM start, interval-heavy
- After 8 weeks: "Sprinter + early-morning regular. Z5 reaches are above average. A deliberate weekly recovery session is recommended."
- Medals: 30 pre-dawn sessions (gold), 3-hour Z4 accumulation (gold)
Scenario C — Travels often, swims occasionally
- Mostly hotel pools while on the road, less than once a week
- After 12 weeks: "Freestyle-leaning free pattern. Your strength is holding a consistent pace."
- Medals: cumulative distance bronze, varied-hour entry bronze
6. Common Questions
Q. My coach notes always read the same. A. When your workouts settle into a steady pattern, the notes settle with them. Vary your strokes, intensity, or session times and the tone will shift.
Q. The note seems wrong. Today was brutal, but it called it a "recovery session." A. FINNS reads quantitative signals (HR zones, pace, SWOLF) rather than subjective fatigue. A wide gap between the numbers and how you felt can itself be a useful signal — possibly a sign to check your conditioning.
Q. Can I pick my persona manually? A. Yes. Settings → AI Coach → "Choose persona manually" — pick one of the 12 and pin it.
Q. Is the coach available in languages other than English? A. iOS supports 11 languages; Android supports 10. The full matrix is in the next post (Part 6).
7. Next Up
What exactly does the AI coach diagnose across those eight areas, and which drills does it prescribe? → Part 11: AI Coach 8 Diagnostic Areas & Try-This Drills
The next post is the heart of the series: the iOS / Android platform feature matrix. A Galaxy Watch user's guide, Samsung Health permission setup, and the actual data differences between the two platforms — all in one place.
App Store / Google Play search: FINNS swim, SwimLog