DEV Community

Run_Dev
Run_Dev

Posted on

How I built a self-adapting training plan in Next.js (no AI, just logic)

#algorithms #javascript #nextjs #showdev

How I built a self-adapting training plan in Next.js (no AI, just logic)

When I started building RunAdapt, I had one goal: a running plan that doesn't fall apart when you miss sessions.

No machine learning. No LLMs. Just deterministic logic that runs every time you mark a session as MISSED or PARTIAL.

Here's exactly how it works.

The core idea

Most training apps are static. You get a plan, and if life gets in the way, you're on your own. RunAdapt runs 4 rules in sequence every time you miss a session.

Rule 1 — Reschedule

Finds the next available slot in your calendar and creates a new session there.

if (session.status === 'MISSED') {
  const nextSlot = findNextAvailableSlot(plan.availableDays);
  createSession({ ...session, date: nextSlot });
}
Enter fullscreen mode Exit fullscreen mode

Rule 2 — Reduce intensity

If you have 2 or more consecutive missed sessions, the intensity factor drops by 0.15 (minimum 0.60). No point pushing hard if you're already struggling.

if (consecutiveMisses >= 2) {
  plan.intensityFactor = Math.max(0.60, plan.intensityFactor - 0.15);
}
Enter fullscreen mode Exit fullscreen mode

Rule 3 — Adjust target date

If your failure rate exceeds 20%, the race date moves forward up to 8 weeks. Better to adjust the goal than to pretend you're on track.

if (failureRate > 0.20) {
  plan.adjustedTargetDate = addWeeks(plan.targetDate, weeksNeeded);
}
Enter fullscreen mode Exit fullscreen mode

Rule 4 — Micro session

If you have less than 30 minutes available, instead of skipping entirely, the plan creates a 20 min / 3 km session. Something is always better than nothing.

if (availableMinutes < 30) {
  createMicroSession({ duration: 20, distance: 3 });
}
Enter fullscreen mode Exit fullscreen mode

Why deterministic logic instead of AI?

I considered using an LLM to make the adaptation smarter. But for this use case, deterministic rules have real advantages:

  • Predictable — the user knows exactly what will happen
  • Fast — no API calls, no latency
  • Cheap — no token costs
  • Trustworthy — no hallucinations

Sometimes simple is better.

Stack

  • Next.js 14 (App Router)
  • Supabase (auth + database)
  • Vercel Hobby (free)

What's next

I'm currently looking for my first real users to validate whether these 4 rules are actually enough — or if I'm missing something obvious.

If you're a runner, I'd love your feedback: runadapt.vercel.app

Top comments (0)