# AI Learning Quest Path Recommender (FastAPI)

## PURPOSE
The video walks through "quest paths" a beginner can take to earn the Tutorial Island
starter pack — Google Essentials, the new Google AI Professional Certificate, the
Andrew Ng / deeplearning.ai track, Elements of AI from the University of Helsinki, the
Anthropic Claude path, the free OpenAI foundations courses, the Kaggle 5-day GenAI
intensive, and the Microsoft/LinkedIn Career Essentials. It then admits the real problem
out loud: "there's just like so many resources and options… paradox of choice."
This service solves that: a learner posts their profile (hours per week, monthly budget,
format preference, vendor affinity) and gets a ranked, explainable quest log.

## WHY THIS MATTERS
It teaches the video's core framing — AI learning is not a straight line, it is a map
where everybody clears Tutorial Island first, then picks a class. The recommender encodes
the actual constraints the presenter names: the €/$49-per-month Coursera subscription
with a 7-day trial, the "sub 10 hours" Google Essentials combo, the 13-hour Google AI
Professional Certificate across eight courses, free-to-audit Andrew Ng videos, and the
50-hour Elements of AI. Learners see that budget and time really do change the answer.

## HOW TO BUILD
Stack: **FastAPI + Pydantic + uvicorn**. Define `LearnerProfile` and `Recommendation`
Pydantic models, hold the catalog as a list of dicts, then rank with a deterministic
`score_path` function (style match, budget gate, vendor affinity). The key idea is
keeping the ranking function swappable for an LLM call later:

```python
def score_path(path, profile):
    score = 100.0
    if path["style"] == profile.learning_style: score += 25
    if path["cost_usd_month"] > profile.budget_usd_month: score -= 40
    # TODO: replace with the description's "pick your class" prompt call
    return Recommendation(path_id=path["id"], fit_score=round(score, 1), ...)
```

## WHEN TO USE IN PRODUCTION
Use it for onboarding funnels at academies, L&D portals, or community sites that want to
route learners to the right course bundle instead of the loudest one. Use it internally
to justify a training budget: the ranked output is a defensible artifact.
Do NOT use it when the catalog is tiny (a hardcoded list beats a service), when you need
licensed curriculum advice, or when the recommendation must reflect real-time enrollment
data — that needs a live upstream API, not a static table.

## HOW IT HELPS STUDENTS
Students learn request/response modeling, input validation with Pydantic, deterministic
ranking logic, and how to leave a clean seam for an LLM/agent call. It is the "web app"
half of production readiness without any ML complexity.

## STARTER CODE
See `starter/main.py` for the runnable FastAPI skeleton (`/starter-pack`, `/paths`,
`/recommend`) with a deterministic scoring stub and `# TODO` markers where the real
LLM ranking and persistence go.
