Social Coach

LLM-powered social coaching system that extracts actionable social skills from books and applies them to real-world scenarios.

The pipeline scrapes books from multiple public sources, extracts social coaching concepts using LLMs, tags and validates them, then refines the final corpus for downstream use.

Book Sources → Scraper → Extractor → Tagger → Validator → Refiner → Final Corpus

1. Scraping: Collects book content from Gutenberg, Open Library, Internet Archive, Google Books
2. Extraction: Uses LLMs to extract social coaching concepts from book text
3. Tagging: Categorizes extracted concepts
4. Validation: Filters and validates quality
5. Refinement: Produces the final curated corpus

src/
  book_scraper.py           # Multi-source book scraper
  archive_client.py         # Internet Archive client
  google_books_client.py    # Google Books client
  gutenberg_client.py       # Project Gutenberg client
  openlibrary_client.py     # Open Library client
  extractor.py              # LLM-based concept extraction
  tagger.py                 # Concept tagging
  validator.py              # Quality validation
  refiner.py                # Corpus refinement
  utils.py                  # Shared utilities
data/
  books_metadata.csv        # Metadata for selected books
  books_selected_200.csv    # 200 selected books
  books_content/            # Local-only downloaded book texts
  corpus_extracted.json     # Raw extracted concepts
  corpus_tagged.json        # Tagged concepts
  corpus_validated.json     # Validated concepts
  corpus_final.json         # Final curated corpus
paper/                      # Manuscript submodule
artifact/                   # Anonymous review artifact submodule

The manuscript is maintained in the separate social-coach-paper repository and included here at paper/ as a Git submodule. The anonymized reproduction package is maintained at artifact/. Clone the project and both submodules with:

git clone --recurse-submodules https://github.com/GeminiLight/social-coach-dev.git

For an existing checkout, initialize or refresh both submodules with:

git submodule update --init --recursive

Raw downloaded book texts and run logs are intentionally kept local and are not committed to the public development repository.

The polished paper homepage is in docs/ and can be served directly by GitHub Pages. It includes the static HTML/CSS/JS site, paper PDF, visual assets, citation metadata, structured data, sitemap.xml, robots.txt, and llms.txt for SEO/GEO.

Local preview:

python3 -m http.server 4173 --directory docs

python src/book_scraper.py      # Scrape books
python src/extractor.py         # Extract concepts
python src/tagger.py            # Tag concepts
python src/validator.py         # Validate
python src/refiner.py           # Refine final corpus

Generate a readable, validated catalog from the selected 200-book CSV:

python3 scripts/export_book_list.py