resume-parser-poc/run.py at main · ethanyandev/resume-parser-poc · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
#!/usr/bin/env python3
"""CLI for the resume extraction POC.

  python run.py path/to/resume.pdf            # run all steps (auto-OCR if scanned) -> HTML
  python run.py path/to/resume.pdf --open     # ...and open the HTML in your browser
  python run.py path/to/resume.pdf --step 1   # text extraction + page render (no LLM)
  python run.py --doc-id <id> --step ocr      # transcribe a scanned PDF's pages
  python run.py --doc-id <id> --step 2        # replay LLM extraction from step1/ocr
  python run.py --doc-id <id> --step 3        # replay verification (no LLM)
  python run.py --doc-id <id> --step html     # re-render the HTML (no LLM)

Artifacts land in artifacts/<doc_id>/ (resume.html is the human-readable output).
AI call trail in artifacts/ai_calls.jsonl.
"""

from __future__ import annotations

import argparse
import json
import subprocess

from dotenv import load_dotenv

from resume_poc import pipeline

load_dotenv()


def main() -> None:
    ap = argparse.ArgumentParser()
    ap.add_argument("pdf", nargs="?", help="path to resume PDF")
    ap.add_argument("--doc-id", help="replay a step from saved artifacts")
    ap.add_argument("--step", choices=["1", "2", "3", "ocr", "html"], help="run a single step")
    ap.add_argument("--no-vision", action="store_true",
                    help="text-only arm (skip page images) for A/B vs the vision path")
    ap.add_argument("--open", dest="open_", action="store_true",
                    help="open the rendered HTML in the default browser")
    args = ap.parse_args()
    use_vision = not args.no_vision

    def show(title: str, obj: dict) -> None:
        print(f"\n=== {title} ===")
        print(json.dumps(obj, ensure_ascii=False, indent=2))

    def doc_id() -> str:
        return args.doc_id or pipeline.doc_id_for(args.pdf)

    def deliver(path: str) -> None:
        print(f"\nHTML -> {path}")
        if args.open_:
            subprocess.run(["open", path], check=False)

    if args.step == "1":
        s1 = pipeline.step1(args.pdf)
        show("step1: text", s1)
        if s1.get("source") == "needs_ocr":
            print(f"\n[scanned] no text layer -> run:  --doc-id "
                  f"{pipeline.doc_id_for(args.pdf)} --step ocr")
    elif args.step == "ocr":
        show("ocr: transcription", pipeline.step_ocr(doc_id()))
    elif args.step == "2":
        show("step2: fields", pipeline.step2(doc_id(), use_vision=use_vision))
    elif args.step == "3":
        show("step3: verify", pipeline.step3(doc_id()))
    elif args.step == "html":
        deliver(pipeline.step_html(doc_id()))
    else:
        out = pipeline.run_all(args.pdf, use_vision=use_vision)
        print(f"\ndoc_id: {out['doc_id']}")
        if "ocr" in out:
            o = out["ocr"]
            print(f"[scanned] OCR via {o['provider']}: {o['n_pages']} page(s), "
                  f"${o['cost_usd']:.5f}")
        show("step2: fields", out["step2"])
        show("step3: verify", out["step3"])
        rep = out["step3"]
        s2 = out["step2"]
        print(f"\nsource: {s2['source']}  mode: {s2['mode']} ({s2['n_images']} page image(s))")
        print(f"confidence: {rep['confidence']}  "
              f"({rep['verified']}/{rep['total_checks']} grounded)")
        if rep["unverified"]:
            print("UNVERIFIED (possible hallucination / 张冠李戴):")
            for c in rep["unverified"]:
                print(f"  - [{c['kind']}] {c['label']}")
        deliver(out["html"])


if __name__ == "__main__":
    main()