| """ |
| Curator pass — one LLM call per saved page. |
| |
| Classifies a page as a "lesson" (French class notes/vocab/grammar) or a |
| "resource" (links, apps, book recommendations) and extracts a friendly |
| title, a one-line summary, and — for resource pages — structured links/books |
| for the Resources tab. |
| """ |
| import logging |
| import llm |
| import prompts |
|
|
| logger = logging.getLogger(__name__) |
|
|
|
|
| def _fallback(raw_text: str) -> dict: |
| first_line = raw_text.strip().split("\n", 1)[0].lstrip("#").strip() |
| return { |
| "title": (first_line or "Untitled Lesson")[:80], |
| "summary": "", |
| "page_type": "lesson", |
| "links": [], |
| "books": [], |
| } |
|
|
|
|
| def curate_page(raw_text: str) -> dict: |
| """Return {title, summary, page_type, links, books} for a page's raw text.""" |
| fallback = _fallback(raw_text) |
| result = llm.chat_json(prompts.CURATOR_SYSTEM, raw_text[:2000], fallback=fallback) |
|
|
| title = (result.get("title") or fallback["title"]).strip()[:80] |
| summary = (result.get("summary") or "").strip() |
| page_type = result.get("page_type") if result.get("page_type") in ("lesson", "resource") else "lesson" |
|
|
| links, books = [], [] |
| if page_type == "resource": |
| for link in result.get("links") or []: |
| if isinstance(link, dict) and link.get("url"): |
| links.append({ |
| "label": str(link.get("label") or link["url"])[:120], |
| "url": str(link["url"]), |
| }) |
| for book in result.get("books") or []: |
| if isinstance(book, dict) and book.get("title"): |
| books.append({ |
| "title": str(book.get("title"))[:200], |
| "author": str(book.get("author") or "")[:120], |
| "note": str(book.get("note") or "")[:200], |
| }) |
|
|
| return { |
| "title": title, |
| "summary": summary, |
| "page_type": page_type, |
| "links": links, |
| "books": books, |
| } |
|
|
