Upload 3 files

README.md

@@ -1,15 +1,27 @@
 ---
-title: Avps
-emoji: 🌖
-colorFrom: red
-colorTo: purple
+title: AVPS OPT-NC
+emoji: 📋
+colorFrom: blue
+colorTo: blue
 sdk: gradio
-sdk_version: 6.14.0
-python_version: '3.13'
+sdk_version: "4.44.0"
 app_file: app.py
 pinned: false
 license: cc-by-4.0
-short_description: Application de recherche augmentée pour les AVPs de l'OPT-NC
+datasets:
+  - opt-nc/avps
+tags:
+  - nouvelle-calédonie
+  - emploi
+  - job-posting
+  - opt-nc
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# AVPS OPT-NC — Avis de Vacances de Poste
+
+Application de recherche et d'exploration des offres d'emploi de l'OPT-NC.
+
+- **Recherche** : recherche libre par mots-clés avec filtres (direction, grade, disponibilité)
+- **Exploration** : statistiques et répartition des offres
+
+Source : [opt-nc/avps](https://huggingface.co/datasets/opt-nc/avps) · Licence CC-BY-4.0

app.py

@@ -0,0 +1,168 @@
+import os
+import gradio as gr
+import pandas as pd
+import numpy as np
+import requests
+from datasets import load_dataset
+
+# --- Chargement des données ---
+ds = load_dataset("opt-nc/avps", split="train")
+df = ds.to_pandas()
+
+embeddings_matrix = np.array(df["embedding"].tolist(), dtype=np.float32)
+norms = np.linalg.norm(embeddings_matrix, axis=1, keepdims=True)
+embeddings_norm = embeddings_matrix / (norms + 1e-10)
+
+EMBED_API = "https://api-inference.huggingface.co/models/BAAI/bge-m3"
+HF_TOKEN = os.environ.get("HF_TOKEN", "")
+
+
+def encode_query(text: str):
+    headers = {"Content-Type": "application/json"}
+    if HF_TOKEN:
+        headers["Authorization"] = f"Bearer {HF_TOKEN}"
+    try:
+        r = requests.post(
+            EMBED_API,
+            headers=headers,
+            json={"inputs": text, "options": {"wait_for_model": True}},
+            timeout=30,
+        )
+        if r.status_code in (429, 503):
+            return None
+        r.raise_for_status()
+        vec = np.array(r.json(), dtype=np.float32)
+        if vec.ndim == 2:
+            vec = vec[0]
+        return vec / (np.linalg.norm(vec) + 1e-10)
+    except Exception:
+        return None
+
+
+def render_cards(results_df, scores=None):
+    if results_df.empty:
+        return "<p style='color:#888;text-align:center;padding:2rem'>Aucun résultat trouvé.</p>"
+
+    cards = []
+    for i, (_, row) in enumerate(results_df.iterrows()):
+        titre = row.get("titre") or "Poste sans titre"
+        direction = row.get("direction_interne") or ""
+        grade = row.get("corps_grade") or ""
+        lieu = row.get("lieu_travail") or ""
+        immed = row.get("disponible_immediatement", False)
+        url = row.get("url") or ""
+        texte = (row.get("text") or "")[:200].strip()
+        score = scores[i] if scores is not None else None
+
+        immed_badge = (
+            '<span style="background:#d1fae5;color:#065f46;font-size:11px;'
+            'padding:2px 8px;border-radius:12px;font-weight:500">⚡ Immédiat</span>'
+            if immed else ""
+        )
+        score_badge = (
+            f'<span style="background:#ede9fe;color:#5b21b6;font-size:11px;'
+            f'padding:2px 8px;border-radius:12px">{score:.0%}</span>'
+            if score is not None else ""
+        )
+        meta_parts = [p for p in [direction, grade, lieu] if p]
+        meta = " · ".join(meta_parts)
+
+        link_btn = (
+            f'<a href="{url}" target="_blank" style="display:inline-block;margin-top:10px;'
+            f'padding:8px 16px;background:#2563eb;color:#fff;border-radius:8px;'
+            f'text-decoration:none;font-size:13px;font-weight:500">Voir l\'annonce →</a>'
+            if url else ""
+        )
+
+        card = f"""
+<div style="background:#fff;border:1px solid #e5e7eb;border-radius:12px;
+            padding:14px 16px;margin-bottom:12px">
+  <div style="display:flex;justify-content:space-between;align-items:flex-start;
+              flex-wrap:wrap;gap:6px;margin-bottom:6px">
+    <span style="font-size:15px;font-weight:600;color:#111;line-height:1.3;
+                 flex:1;min-width:0">{titre}</span>
+    <div style="display:flex;gap:4px;flex-shrink:0">{score_badge}{immed_badge}</div>
+  </div>
+  <p style="font-size:12px;color:#6b7280;margin:0 0 6px">{meta}</p>
+  <p style="font-size:13px;color:#374151;margin:0;line-height:1.5">{texte}…</p>
+  {link_btn}
+</div>"""
+        cards.append(card)
+
+    header = f'<p style="font-size:13px;color:#6b7280;margin-bottom:12px">{len(results_df)} offre(s) trouvée(s)</p>'
+    return header + "\n".join(cards)
+
+
+def search(query):
+    if not query.strip():
+        # Afficher toutes les offres par défaut
+        return render_cards(df)
+
+    q_vec = encode_query(query)
+
+    if q_vec is not None:
+        sims = embeddings_norm @ q_vec
+        order = np.argsort(sims)[::-1]
+        results = df.iloc[order].copy()
+        scores = sims[order].tolist()
+        # Garder seulement les résultats pertinents (similarité > 0.3)
+        mask = [s > 0.3 for s in scores]
+        results = results[mask]
+        scores = [s for s, m in zip(scores, mask) if m]
+        return render_cards(results, scores)
+    else:
+        # Fallback mots-clés
+        qwords = query.lower().split()
+        def score_row(row):
+            text = f"{row.get('titre','')} {row.get('text','')} {row.get('corps_grade','')}".lower()
+            return sum(1 for w in qwords if w in text)
+        df2 = df.copy()
+        df2["_s"] = df2.apply(score_row, axis=1)
+        df2 = df2[df2["_s"] > 0].sort_values("_s", ascending=False)
+        return render_cards(df2)
+
+
+# --- CSS global pour mobile ---
+css = """
+body { max-width: 600px; margin: 0 auto; }
+.gradio-container { padding: 0 !important; }
+footer { display: none !important; }
+#query textarea {
+    font-size: 16px !important;
+    line-height: 1.5 !important;
+}
+#search-btn {
+    font-size: 16px !important;
+    height: 48px !important;
+    border-radius: 10px !important;
+}
+"""
+
+with gr.Blocks(css=css, title="AVPS OPT-NC") as demo:
+    gr.Markdown(
+        "## 📋 Offres OPT-NC\nDécrivez votre profil ou saisissez des mots-clés.",
+    )
+
+    query_input = gr.Textbox(
+        elem_id="query",
+        label="",
+        placeholder=(
+            "Ex : \"Cadre avec expérience en gestion de projets SI et management d'équipe\"\n"
+            "ou : chef de service, télécoms, RH, marketing..."
+        ),
+        lines=4,
+        max_lines=8,
+        show_label=False,
+    )
+
+    search_btn = gr.Button("🔍 Rechercher", variant="primary", elem_id="search-btn")
+
+    results_html = gr.HTML()
+
+    # Lancement au démarrage
+    demo.load(fn=search, inputs=query_input, outputs=results_html)
+    search_btn.click(fn=search, inputs=query_input, outputs=results_html)
+    query_input.submit(fn=search, inputs=query_input, outputs=results_html)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,4 @@
+gradio>=4.0.0
+datasets
+pandas
+numpy