How the AI training recommendations work

Below your progress bars, the AI Next steps panel suggests what to focus on at your current stage. It's not a replacement for your CFI's judgment — it's a CFI-style heuristic layered on top of the deterministic progress data.

What it actually does

The system runs in two layers:

1. Rules engine (always runs, free, deterministic). Calculates progress percentages, hours remaining, training velocity, completion-date estimates.
2. LLM layer (Llama 3.3 70B by default). Builds a CFI-persona prompt with your training data and returns:
   • Next-step recommendations with reasoning
   • Focus area analysis — which requirements are lagging or accelerating
   • Session planning — what to fly next
   • Pattern recognition — gaps in your log (no recent night flying, no solo cross-country yet)
   • Encouragement — qualitative summary

Why it's cached

The LLM call is expensive. To keep the experience fast and the cost manageable:

• We compute a fingerprint of your training data (logbook hash + endorsement count + license ID + progress hash).
• If the fingerprint hasn't changed since the last call, we return the cached response — zero LLM call.
• The cache lives for up to 7 days even if nothing changed, just to avoid surfacing stale advice.

In practice this means logging a new flight or getting a new endorsement triggers a fresh AI run; refreshing the page without changes doesn't.

When it falls back

If the LLM is unavailable (network issue, rate limit, missing API key), the panel quietly drops to rules-only mode. You'll still see the progress numbers, just without the qualitative recommendations. There's no error toast — the absence of AI shouldn't break your day.

Provider transparency

The LLM provider is configurable per environment by the platform team. The default is Groq's hosted Llama 3.3 70B because it's free at our usage tier and fast enough to feel synchronous. Schools on enterprise contracts can request a different provider or an on-prem deployment.

What it never does

• The AI doesn't write or sign anything in your logbook.
• It doesn't issue endorsements.
• It doesn't talk to the FAA.
• It doesn't make scheduling decisions on your behalf — only your CFI or a school admin does that.