backend/seed.py

# backend/seed.py
"""
Build the demo index at Docker-image build time.

Parses every PDF in data/uploads (the committed public demo filing) into
the graph + Chroma store so a fresh container boots ready to answer —
no 3-minute first-upload wait in front of a recruiter, and no 60MB+ of
binary index files in git.

Run from the repo root:  python -m backend.seed
Idempotent: skips work if the graph file already exists.
"""

from pathlib import Path

from backend.graph import FinancialGraph
from backend.parser import parse_document

DATA_DIR = Path("data")
GRAPH_PATH = DATA_DIR / "graph.json"
UPLOADS_DIR = DATA_DIR / "uploads"

# company/year for known demo filings; anything unlisted falls back to
# a name derived from the file name
KNOWN_FILINGS = {
    "HDFC_Bank_Annual_Report_2024_25-310202.pdf": ("HDFC Bank", "2025"),
}


def main():
    if GRAPH_PATH.exists():
        print(f"seed: {GRAPH_PATH} already exists, skipping")
        return

    pdfs = sorted(UPLOADS_DIR.glob("*.pdf")) if UPLOADS_DIR.is_dir() else []
    if not pdfs:
        print("seed: no PDFs in data/uploads, nothing to do")
        return

    fg = FinancialGraph()
    for pdf in pdfs:
        company, year = KNOWN_FILINGS.get(
            pdf.name, (pdf.stem.replace("_", " ")[:40], "2025")
        )
        print(f"seed: parsing {pdf.name} as {company} {year}")
        parsed = parse_document(str(pdf), company, year)
        # pattern-only relations: no LLM available (or wanted) at build time
        fg.add_document(parsed, use_llm_fallback=False)

    DATA_DIR.mkdir(exist_ok=True)
    fg.save(str(GRAPH_PATH))
    print("seed: done")


if __name__ == "__main__":
    main()