Download app.py from Imene-ee/LabLens: direct link, hf CLI and curl.
- Browser
- Download file 9.82 kB
-
https://huggingface.co/spaces/Imene-ee/LabLens/resolve/main/app.py
- Command line
-
hf download hf://spaces/Imene-ee/LabLens/app.py
-
curl -L -o app.py https://huggingface.co/spaces/Imene-ee/LabLens/resolve/main/app.py
9.82 kB
| """LabLens web app (Gradio). Runs locally or as a Hugging Face Space. | |
| Flow: upload -> read (deterministic extraction + flags) -> you check the table -> explain. | |
| The check step is deliberate: OCR can misread a digit, and a human glance is the cheapest | |
| safety net there is. | |
| """ | |
| from __future__ import annotations | |
| import html | |
| from pathlib import Path | |
| import gradio as gr | |
| from lablens.catalog import load_catalog, match_test_name | |
| from lablens.explain import explain, llm_available, render_markdown | |
| from lablens.extract import extract | |
| from lablens.flagger import flag_all | |
| from lablens.i18n import CHECK_NOTE, DISCLAIMER, FLAG_LABEL, GENERIC_RANGE, LANGS, RTL, URGENT, t | |
| from lablens.models import Flag, Result | |
| # ZeroGPU Spaces stop any app that declares no GPU function at startup. LabLens runs on | |
| # CPU, so declare a no-op one. Outside Hugging Face (no `spaces` package) this is skipped. | |
| try: | |
| import spaces | |
| def _zerogpu_keepalive(): | |
| return None | |
| except ImportError: | |
| pass | |
| SAMPLES = Path(__file__).parent / "samples" | |
| HEADERS = ["Test", "Value", "Unit", "Range low", "Range high"] | |
| LANG_CHOICES = [(v, k) for k, v in LANGS.items()] | |
| catalog = load_catalog() | |
| FLAG_COLOR = { | |
| "normal": "ok", "low": "warn", "high": "warn", "critical_low": "crit", "critical_high": "crit", | |
| "unknown": "muted", "check": "check", | |
| } | |
| def _num(x): | |
| if x is None: | |
| return None | |
| s = str(x).strip().replace(",", ".") | |
| if s in ("", "nan", "None", "-"): | |
| return None | |
| try: | |
| return float(s) | |
| except ValueError: | |
| return None | |
| def read_report(file, lang): | |
| if file is None: | |
| return gr.update(value=[]), "Upload a report or pick a sample first.", None | |
| path = file if isinstance(file, str) else file.name | |
| results, unmatched, detected, method = extract(path) | |
| analysis = flag_all(results, method=method) | |
| rows = [[catalog[r.test_key].display["en"], r.value, r.unit or "", r.ref_low, r.ref_high] | |
| for r in analysis.results] | |
| n_check = sum(r.flag == Flag.CHECK for r in analysis.results) | |
| how = "the PDF's text" if method == "pdf-text" else "the photo (OCR)" | |
| status = (f"Read **{len(rows)} results** from {how}. " | |
| "Compare the table with your report, fix anything that's wrong, then press **Explain**.") | |
| if n_check: | |
| status += f"\n\n⚠️ {t(CHECK_NOTE, lang)}" | |
| if unmatched: | |
| status += f"\n\n<details><summary>{len(unmatched)} line(s) with numbers I couldn't match to a test</summary>\n\n" + \ | |
| "\n".join(f"- `{html.escape(u)}`" for u in unmatched[:15]) + "\n</details>" | |
| return gr.update(value=rows), status, detected | |
| def _rows_to_results(rows) -> list[Result]: | |
| out = [] | |
| data = rows.values.tolist() if hasattr(rows, "values") else (rows or []) | |
| for row in data: | |
| if not row or not str(row[0]).strip(): | |
| continue | |
| key, score = match_test_name(str(row[0])) | |
| value = _num(row[1]) | |
| if key is None or value is None: | |
| continue | |
| lo, hi = _num(row[3]), _num(row[4]) | |
| out.append(Result(test_key=key, raw_name=str(row[0]), value=value, unit=str(row[2] or "") or None, | |
| ref_low=lo, ref_high=hi, | |
| ref_source="report" if (lo is not None or hi is not None) else "none", | |
| match_score=score)) | |
| return out | |
| def _results_table(analysis, lang) -> str: | |
| rtl = ' dir="rtl"' if lang in RTL else "" | |
| trs = [] | |
| for r in analysis.results: | |
| td = catalog[r.test_key] | |
| name = html.escape(td.display.get(lang) or td.display["en"]) | |
| if r.ref_low is not None and r.ref_high is not None: | |
| rng = f"{r.ref_low:g} – {r.ref_high:g}" | |
| elif r.ref_high is not None: | |
| rng = f"< {r.ref_high:g}" | |
| elif r.ref_low is not None: | |
| rng = f"> {r.ref_low:g}" | |
| else: | |
| rng = "—" | |
| if r.ref_source == "catalog": | |
| rng += f'<div class="ll-note">{html.escape(t(GENERIC_RANGE, lang))}</div>' | |
| label = html.escape(t(FLAG_LABEL[r.flag.value], lang)) | |
| trs.append(f'<tr><td>{name}</td><td class="ll-num">{r.value:g} <span class="ll-unit">{html.escape(r.unit or "")}</span></td>' | |
| f'<td class="ll-num">{rng}</td><td><span class="ll-badge ll-{FLAG_COLOR[r.flag.value]}">{label}</span></td></tr>') | |
| return f'<table class="ll-table"{rtl}><tbody>{"".join(trs)}</tbody></table>' | |
| def run_explain(rows, lang): | |
| results = _rows_to_results(rows) | |
| if not results: | |
| return "", "", "Nothing to explain yet: read a report first.", "" | |
| analysis = flag_all(results) | |
| rtl = ' dir="rtl"' if lang in RTL else "" | |
| banner = "" | |
| if analysis.urgent: | |
| banner = f'<div class="ll-urgent"{rtl}>🚨 {_md_bold(t(URGENT, lang))}</div>' | |
| if any(r.flag == Flag.CHECK for r in analysis.results): | |
| banner += f'<div class="ll-check"{rtl}>⚠️ {html.escape(t(CHECK_NOTE, lang))}</div>' | |
| exp, source = explain(analysis, lang) | |
| md = render_markdown(analysis, exp, lang) | |
| if source == "llm": | |
| note = "Explanation written by an open model, grounded in the LabLens knowledge base and checked by guardrails." | |
| elif source == "template:no-api-key": | |
| note = "Explanation from the LabLens knowledge base (no AI key configured: add LLM_API_KEY or HF_TOKEN)." | |
| else: | |
| note = ("Explanation from the LabLens knowledge base (the AI call failed: " | |
| f"{html.escape(source.removeprefix('template:'))})") | |
| return banner, _results_table(analysis, lang), md, f"<small>{note} {html.escape(t(DISCLAIMER, lang))}</small>" | |
| def _md_bold(s: str) -> str: | |
| s = html.escape(s) | |
| import re | |
| return re.sub(r"\*\*(.+?)\*\*", r"<strong>\1</strong>", s) | |
| def load_sample(name): | |
| return str(SAMPLES / name) | |
| CSS = """ | |
| :root { --ll-ink:#14213d; --ll-accent:#0f766e; } | |
| .gradio-container { max-width: 1080px !important; margin: auto; } | |
| #ll-hero h1 { font-size: 2.1rem; margin-bottom: .2rem; letter-spacing: -0.02em; } | |
| #ll-hero p { font-size: 1.02rem; opacity: .85; margin-top: 0; } | |
| .ll-pill { display:inline-block; padding:.18rem .6rem; border-radius:999px; font-size:.78rem; | |
| background: color-mix(in srgb, var(--ll-accent) 14%, transparent); color: var(--ll-accent); | |
| margin-right:.35rem; font-weight:600; } | |
| .ll-table { width:100%; border-collapse:collapse; font-size:.95rem; } | |
| .ll-table td { padding:.55rem .6rem; border-bottom:1px solid color-mix(in srgb, currentColor 12%, transparent); vertical-align: top; } | |
| .ll-num { font-variant-numeric: tabular-nums; white-space: nowrap; } | |
| .ll-unit { opacity:.65; font-size:.85em; } | |
| .ll-note { font-size:.75rem; opacity:.65; white-space: normal; } | |
| .ll-badge { padding:.15rem .55rem; border-radius:6px; font-size:.82rem; font-weight:600; white-space:nowrap; } | |
| .ll-ok { background:#dcfce7; color:#14532d; } | |
| .ll-warn { background:#fef3c7; color:#78350f; } | |
| .ll-crit { background:#fee2e2; color:#7f1d1d; } | |
| .ll-check { background:#e0e7ff; color:#312e81; padding:.7rem .9rem; border-radius:10px; margin:.4rem 0; } | |
| .ll-muted { background:#e5e7eb; color:#374151; } | |
| .ll-urgent { background:#fee2e2; color:#7f1d1d; border:2px solid #dc2626; padding:.9rem 1rem; | |
| border-radius:12px; font-size:1.02rem; margin:.4rem 0; } | |
| """ | |
| with gr.Blocks(title="LabLens") as demo: | |
| gr.HTML( | |
| '<div id="ll-hero"><h1>🔬 LabLens</h1>' | |
| "<p>Understand your blood test in plain language — English, Français, العربية, or Tounsi.</p>" | |
| '<span class="ll-pill">Flags by fixed medical rules, not AI</span>' | |
| '<span class="ll-pill">Your image and name never reach the AI</span>' | |
| '<span class="ll-pill">You check every value</span></div>' | |
| ) | |
| with gr.Row(): | |
| with gr.Column(scale=5): | |
| file = gr.File(label="Lab report (photo or PDF)", file_types=["image", ".pdf"], type="filepath") | |
| with gr.Row(): | |
| s1 = gr.Button("Sample: English PDF", size="sm") | |
| s2 = gr.Button("Sample: French photo", size="sm") | |
| s3 = gr.Button("Sample: urgent value", size="sm") | |
| lang = gr.Dropdown(LANG_CHOICES, value="en", label="Explain in") | |
| read_btn = gr.Button("1 · Read report", variant="primary") | |
| with gr.Column(scale=7): | |
| status = gr.Markdown("Upload a report to begin.") | |
| table = gr.Dataframe(headers=HEADERS, datatype=["str", "number", "str", "number", "number"], | |
| interactive=True, label="What I read — edit anything that's wrong", | |
| row_count=(0, "dynamic"), column_count=(5, "fixed")) | |
| explain_btn = gr.Button("2 · Explain my results", variant="primary") | |
| detected = gr.State(None) | |
| banner = gr.HTML() | |
| results_html = gr.HTML() | |
| explanation = gr.Markdown() | |
| footnote = gr.HTML(f"<small>{t(DISCLAIMER, 'en')} Synthetic sample reports only — fictional patients and labs.</small>") | |
| s1.click(lambda: load_sample("sample_en_report.pdf"), outputs=file) | |
| s2.click(lambda: (load_sample("sample_fr_photo.png"), "fr"), outputs=[file, lang]) | |
| s3.click(lambda: (load_sample("sample_fr_urgent.png"), "darija"), outputs=[file, lang]) | |
| read_btn.click(read_report, [file, lang], [table, status, detected]) | |
| explain_btn.click(run_explain, [table, lang], [banner, results_html, explanation, footnote]) | |
| lang.change(lambda l: gr.update(rtl=l in RTL), lang, explanation) | |
| if __name__ == "__main__": | |
| import os | |
| print("LLM configured:", llm_available(), "| base:", os.getenv("LLM_BASE_URL", "default (HF router)"), | |
| "| model:", os.getenv("LLM_MODEL", "default"), flush=True) | |
| demo.launch(css=CSS, theme=gr.themes.Soft(primary_hue="teal"), ssr_mode=False) | |