1. Resolve registry URL
   • NPM: https://registry.npmjs.org/{package_name}/{version}
   • PyPI: https://pypi.org/pypi/{name}/{version}/json
2. Fetch README & docs
   • readme = github.get_file("README.md")
   • docs_tree = github.list_dir("/docs")
3. Scrape extra docs via Apify
   • Start URLs: homepage, docs_url.
   • Max pages: 30.
4. StackOverflow Q&A (votes ≥ 10).
   • API: GET /2.3/questions?order=desc&sort=votes&tagged={pkg}&site=stackoverflow
5. Chunking
   • Algorithm: recursive text splitter 800 tokens overlap = 50.
6. Embedding
   • embedding = openai.Embedding.create(input=chunk, model=env.EMBEDDING_MODEL)["data"][0]["embedding"]
7. Supabase insert
   • INSERT INTO documentation (...)
8. Profile summary (Gemini prompt)
   • Sections: Install · Quickstart · API Surface.
9. Cache
   • redis.setex(profile_key, 35*86400, json.dumps(summary))
10. Cold store
    • boto3.put_object(Bucket=S3_BUCKET, Key=f"profiles/{...}.json") (S3 lifecycle: delete after 60 d).