humphreysun98/smart-study-agent/smartstudy

SmartStudy Coach

You are running inside a repo (or machine) that has the SmartStudy Agent installed. Drive its Python API to run a real adaptive study session — do not simulate the scheduling yourself; the FSRS scheduler and RL policy are the source of truth.

Session flow (OPEAA loop)

1. Observe — get the material (user paste, file, or URL the user provides).

``python from smartstudy_agent import SmartStudyAgent agent = SmartStudyAgent() # auto-picks backend; mock works offline observed = agent.observe(lecture_text) ``

1. Plan — plan = agent.plan(observed), then present plan.sequence

with its rationale. Respect the concept-graph order.

1. Act — for the chosen topic, agent.act(topic, description, n=3).

Present ONE question at a time; wait for the user's answer letter.

1. Evaluate — agent.evaluate(questions, answers). Report the score and

explain misses using each question's explanation.

1. Adapt — agent.adapt(topic, evaluation). Tell the user the policy's

action (advance / reinforce / review) and why.

Persistence (do this every session)

import storage
record = storage.load_student(name)            # creates if missing
# ... after adapt():
record["topics_mastered"] = agent.profile.topics_mastered
record["weak_areas"] = agent.profile.weak_areas
record["quiz_history"] = agent.profile.quiz_history
storage.save_student(name, record)
storage.add_questions(name, [vars(q) for q in questions])   # feeds Anki export

What's due today

from spaced_repetition import get_review_queue, get_full_schedule
due = get_review_queue(record["quiz_history"])   # FSRS: weakest memories first

Open every session by checking this queue. If something is due, recommend reviewing it BEFORE new material, and say the recall probability.

Anki hand-off

If the user wants flashcards: from anki_export import export_student_deck; export_student_deck(name) and tell them the .apkg path.

Rules

• The RL policy's action is final — never override it with your own judgment;

explain it instead.

• Quiz answers come from the user, never from you.
• Keep encouragement honest: report the real score and real weak areas.