Update app.py

app.py

@@ -1,31 +1,26 @@
 import os
 from pathlib import Path
+import re
+import io
+import time
+import json
+from collections import Counter
+
 import gradio as gr
 import numpy as np
 import faiss
 from sentence_transformers import SentenceTransformer
 from openai import OpenAI, OpenAIError
 
-# ----------------------------
-# NVIDIA API Key (via Space Variables & Secrets)
-# ----------------------------
+# ========= NVIDIA API =========
 NV_API_KEY = os.environ.get("NV_API_KEY")
 if not NV_API_KEY:
-    raise RuntimeError(
-        "🔒 NV_API_KEY not set. "
-        "Configure it in Settings → Variables & Secrets."
-    )
+    raise RuntimeError("🔒 NV_API_KEY not set. Configure it em Settings → Variables & Secrets.")
 
-# NVIDIA-compatible OpenAI client for chat
-client = OpenAI(
-    base_url="https://integrate.api.nvidia.com/v1",
-    api_key=NV_API_KEY
-)
+client = OpenAI(base_url="https://integrate.api.nvidia.com/v1", api_key=NV_API_KEY)
 CHAT_MODEL = "meta/llama3-8b-instruct"
 
-# ----------------------------
-# Application configuration
-# ----------------------------
+# ========= App config =========
 APP_TITLE = "CVchat – Ronaldo Menezes"
 INTRO = (
     "👋 Olá! Eu sou o CVchat do Ronaldo Menezes.\n"
@@ -45,113 +40,271 @@ SUGGESTION_QUESTIONS = [
     "Certificações?",
 ]
 
-# Paths for FAISS index and chunks
+# (NEW) sugestões por tema
+SUGGESTIONS_THEMES = {
+    "Projetos financiados": [
+        "Liste projetos com financiamento público (CNPq, QREN, UE) e resultados.",
+        "Qual foi o impacto de projetos financiados (KPIs, prazos, orçamento)?",
+    ],
+    "Artigos & Publicações": [
+        "Quais artigos/publicações mais relevantes e onde foram publicados?",
+        "Resumo de publicações sobre sensoriamento remoto (Landsat/Sentinel).",
+    ],
+    "Habilidades técnicas": [
+        "Stack técnica principal (linguagens, libs, cloud, bancos).",
+        "Experiência com FAISS, RAG e LLMs na prática.",
+    ],
+    "Liderança & Gestão": [
+        "Experiência liderando equipes/projetos e responsabilidades.",
+        "Exemplos de melhorias de processo e resultados mensuráveis.",
+    ],
+}
+
+# ========= Paths =========
 INDEX_FILE  = "r_docs.index"
 CHUNKS_FILE = "r_chunks.npy"
 PDF_PATH    = "CV-Ronaldo_Menezes_2025_06.pdf"
 
-# Verify that the index files exist
 if not Path(INDEX_FILE).exists() or not Path(CHUNKS_FILE).exists():
-    raise FileNotFoundError(
-        "Index not found. Run build_index.py to generate r_docs.index and r_chunks.npy."
-    )
+    raise FileNotFoundError("Index not found. Run build_index.py to generate r_docs.index and r_chunks.npy.")
 
-# Load FAISS index and chunks
+# Load FAISS & chunks
 index  = faiss.read_index(INDEX_FILE)
 chunks = np.load(CHUNKS_FILE, allow_pickle=True)
 
-# ----------------------------
-# Context retrieval (local embeddings)
-# ----------------------------
+# ========= Embeddings =========
 embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
 
+# (NEW) pré-cálculo de embedding médio do CV (para match score global)
+_cv_emb_mean = None
+def _ensure_cv_mean():
+    global _cv_emb_mean
+    if _cv_emb_mean is None:
+        embs = embedding_model.encode(list(chunks), convert_to_numpy=True, normalize_embeddings=True)
+        _cv_emb_mean = embs.mean(axis=0)
+    return _cv_emb_mean
+
 def retrieve_context(query: str, k: int = 4) -> str:
     q_emb = embedding_model.encode([query], convert_to_numpy=True)
     _, I = index.search(q_emb, k)
     return "\n---\n".join(chunks[i] for i in I[0])
 
-# ----------------------------
-# Chatbot state
-# ----------------------------
+# ========= Chat state =========
 dialog_history: list[tuple[str, str]] = []
 
-# ----------------------------
-# Chat function
-# ----------------------------
+# ========= Helpers – NVIDIA chat =========
+def nv_stream(messages, temperature, top_p, max_tokens):
+    """Streaming para o chat principal."""
+    assistant_reply = ""
+    stream = client.chat.completions.create(
+        model=CHAT_MODEL, messages=messages,
+        temperature=temperature, top_p=top_p, max_tokens=max_tokens, stream=True
+    )
+    for chunk in stream:
+        delta = chunk.choices[0].delta
+        if hasattr(delta, "content") and delta.content:
+            assistant_reply += delta.content
+            yield assistant_reply
+
+def nv_complete(messages, temperature, top_p, max_tokens) -> str:
+    """Completa de uma vez (para PDFs e utilitários)."""
+    resp = client.chat.completions.create(
+        model=CHAT_MODEL, messages=messages,
+        temperature=temperature, top_p=top_p, max_tokens=max_tokens, stream=False
+    )
+    return resp.choices[0].message.content.strip()
+
+# ========= PDF utils (NEW) =========
+def _to_pdf_bytes(title: str, body: str) -> bytes:
+    # usa um PDF mínimo sem dependências (texto simples em uma “folha”)
+    # para melhor tipografia, você pode trocar por reportlab.
+    from reportlab.pdfgen import canvas
+    from reportlab.lib.pagesizes import A4
+    from reportlab.lib.utils import simpleSplit
+    buf = io.BytesIO()
+    c = canvas.Canvas(buf, pagesize=A4)
+    w, h = A4
+    margin = 50
+    c.setTitle(title)
+    c.setFont("Helvetica-Bold", 14)
+    c.drawString(margin, h - margin, title)
+    c.setFont("Helvetica", 11)
+    y = h - margin - 30
+    lines = simpleSplit(body, "Helvetica", 11, w - 2 * margin)
+    for line in lines:
+        if y < margin:
+            c.showPage()
+            c.setFont("Helvetica", 11)
+            y = h - margin
+        c.drawString(margin, y, line)
+        y -= 15
+    c.showPage()
+    c.save()
+    buf.seek(0)
+    return buf.read()
+
+# ========= Chat principal =========
 def chatbot(user_input: str, temperature: float, top_p: float, max_tokens: int):
     global dialog_history
     if not user_input:
         return dialog_history, ""
 
-    # Retrieve context
     context = retrieve_context(user_input)
     system_msg = {
         "role": "system",
         "content": (
-            "You are an assistant specialized in R packages. "
-            "Use only the context below to answer. If you don't know, say so.\n\n"
+            "You are an assistant specialized in the candidate's CV. "
+            "Use ONLY the retrieved context to answer. If you don't know, say you don't know.\n\n"
             f"=== Retrieved Context ===\n{context}\n\n"
         )
     }
 
-    # Build messages list
     messages = [system_msg]
     for u, a in dialog_history:
-        messages.extend([
-            {"role": "user", "content": u},
-            {"role": "assistant", "content": a}
-        ])
+        messages.extend([{"role": "user", "content": u}, {"role": "assistant", "content": a}])
     messages.append({"role": "user", "content": user_input})
 
-    # Call NVIDIA chat API in streaming mode
-    assistant_reply = ""
+    # Stream ( mantém UX atual )
+    reply_full = ""
     try:
-        stream = client.chat.completions.create(
-            model=CHAT_MODEL,
-            messages=messages,
-            temperature=temperature,
-            top_p=top_p,
-            max_tokens=max_tokens,
-            stream=True
-        )
-        for chunk in stream:
-            delta = chunk.choices[0].delta
-            if hasattr(delta, "content") and delta.content:
-                assistant_reply += delta.content
+        for partial in nv_stream(messages, temperature, top_p, max_tokens):
+            reply_full = partial
+        dialog_history.append((user_input, reply_full))
     except OpenAIError as e:
-        assistant_reply = f"⚠️ API Error: {e.__class__.__name__}: {e}"
+        reply_full = f"⚠️ API Error: {e.__class__.__name__}: {e}"
+        dialog_history.append((user_input, reply_full))
 
-    # Update history and return
-    dialog_history.append((user_input, assistant_reply))
     return dialog_history, ""
 
-# ----------------------------
-# Clear history
-# ----------------------------
 def clear_history():
     global dialog_history
     dialog_history = []
     return [], ""
 
-# ----------------------------
-# Gradio UI
-# ----------------------------
-custom_css = r"""
-:root {
-  --primary: #4a90e2;
-  --accent: #50e3c2;
-  --bg-light: #f9f9f9;
-  --txt-dark: #333333;
-  --radius: 8px;
-  --spacing: 1rem;
+# ========= (NEW) Mini-bio multi-formato =========
+MINI_BIO_STYLES = {
+    "Acadêmico": "Estilo acadêmico, objetivo, cite publicações/projetos e área de pesquisa.",
+    "Corporativo": "Tom profissional para negócios, destaque resultados, KPIs e liderança.",
+    "Pitch curto": "3-4 frases diretas, chamando atenção para conquistas-chave."
 }
+
+def generate_mini_bio(style_key: str, temperature: float, top_p: float, max_tokens: int):
+    if style_key not in MINI_BIO_STYLES:
+        return None, "Selecione um formato de mini-bio."
+    # usa os top contextos “genéricos”: consulta neutra para cobrir o CV
+    context = retrieve_context("resumo do currículo, principais resultados e tecnologias", k=8)
+    system_msg = {
+        "role": "system",
+        "content": (
+            "Use apenas o contexto do CV para gerar uma mini-bio. "
+            "Não invente fatos. Seja fiel ao conteúdo.\n\n"
+            f"=== Contexto do CV ===\n{context}\n"
+        )
+    }
+    user_msg = {
+        "role": "user",
+        "content": f"Produza uma mini-bio em português. Estilo: {MINI_BIO_STYLES[style_key]} (150-220 palavras)."
+    }
+    try:
+        text = nv_complete([system_msg, user_msg], temperature, top_p, max_tokens)
+        pdf_bytes = _to_pdf_bytes(f"Mini-bio ({style_key})", text)
+        filename = f"mini_bio_{style_key.replace(' ','_').lower()}_{int(time.time())}.pdf"
+        with open(filename, "wb") as f:
+            f.write(pdf_bytes)
+        return filename, "Mini-bio gerada com sucesso."
+    except OpenAIError as e:
+        return None, f"⚠️ API Error: {e}"
+
+# ========= (NEW) Carta de motivação + Match score =========
+def generate_cover_letter(job_desc: str, temperature: float, top_p: float, max_tokens: int):
+    if not job_desc or not job_desc.strip():
+        return None, "Cole a descrição da vaga primeiro."
+    context = retrieve_context(job_desc, k=8)
+    sys = {
+        "role": "system",
+        "content": (
+            "Gere uma carta de motivação baseada SOMENTE no CV (contexto) e na vaga. "
+            "Inclua 2-3 conquistas mensuráveis e tecnologias relevantes. 250-350 palavras.\n\n"
+            f"=== Contexto (CV) ===\n{context}\n"
+        )
+    }
+    usr = {"role": "user", "content": f"Descrição da vaga:\n{job_desc}\n\nGerar carta em PT-BR/PT-PT, tom profissional."}
+    try:
+        text = nv_complete([sys, usr], temperature, top_p, max_tokens)
+        pdf_bytes = _to_pdf_bytes("Carta de Motivação", text)
+        filename = f"carta_{int(time.time())}.pdf"
+        with open(filename, "wb") as f:
+            f.write(pdf_bytes)
+        return filename, "Carta gerada com sucesso."
+    except OpenAIError as e:
+        return None, f"⚠️ API Error: {e}"
+
+def compute_match_score(job_desc: str):
+    """
+    Score 0-100 = 60% similaridade (job vs CV médio) + 40% cobertura de requisitos.
+    Requisitos = palavras-chave (simples) extraídas da vaga; cobertura = % presentes no contexto recuperado.
+    """
+    if not job_desc or not job_desc.strip():
+        return "Cole a descrição da vaga para calcular o match score."
+
+    # Similaridade global
+    cv_mean = _ensure_cv_mean()
+    job_emb = embedding_model.encode([job_desc], convert_to_numpy=True, normalize_embeddings=True)[0]
+    sim = float(np.dot(cv_mean, job_emb))  # [-1,1] com embs normalizados
+    sim_norm = max(0.0, min(1.0, (sim + 1) / 2))  # → [0,1]
+
+    # Requisitos/cobertura (heurística simples)
+    req_tokens = re.findall(r"[a-zA-ZÀ-ÿ0-9\-\+#\.]{3,}", job_desc.lower())
+    stop = set(["com","para","dos","das","uma","um","de","da","do","and","the","with","sem","em","na","no","os","as","que"])
+    req_keywords = [t for t in req_tokens if t not in stop]
+    most_common = [w for w, _ in Counter(req_keywords).most_common(20)]
+
+    retrieved = retrieve_context(job_desc, k=8).lower()
+    hits = sum(1 for w in most_common if w in retrieved)
+    coverage = hits / max(1, len(most_common))  # [0,1]
+
+    score = int(round(100 * (0.6 * sim_norm + 0.4 * coverage)))
+    explain = (
+        f"Similaridade global: {int(sim_norm*100)}% | "
+        f"Cobertura de requisitos: {int(coverage*100)}% | "
+        f"→ Match score: **{score}/100**"
+    )
+    return explain
+
+# ========= (NEW) Métricas do CV =========
+TECH_HINTS = ["python","r","faiss","qdrant","pytorch","tensorflow","scikit","gradio","streamlit",
+              "gis","qgis","gdal","grass","sentinel","landsat","process mining","rag","vit","mask2former"]
+
+COUNTRY_HINTS = ["portugal","brasil","germany","alemanh", "spain","espanha","europe","europa","france","italy","uk","usa"]
+
+def extract_metrics():
+    text_all = " \n".join(map(str, chunks))
+    # Publicações aproximadas: procura palavras-chave
+    pubs = len(re.findall(r"\b(publica(?:ç(?:ões|ao|ão)|dos?)|paper|article|artigo|ieee|springer|acm)\b", text_all, flags=re.I))
+    # Anos presentes (para dar noção de linha do tempo)
+    years = sorted(set(re.findall(r"\b(20\d{2}|19\d{2})\b", text_all)))
+    # Tecnologias (contagem simples)
+    tech_counts = {t: len(re.findall(re.escape(t), text_all, flags=re.I)) for t in TECH_HINTS}
+    top_tech = sorted([k for k,v in tech_counts.items() if v>0], key=lambda k: tech_counts[k], reverse=True)[:8]
+    # Presença internacional
+    intl_hits = sum(len(re.findall(c, text_all, flags=re.I)) for c in COUNTRY_HINTS)
+
+    md = [
+        "### Métricas do CV (estimativas)\n",
+        f"- **Publicações (sinalizadas)**: ~{pubs}",
+        f"- **Anos mencionados**: {', '.join(years[:12])}{'…' if len(years)>12 else ''}",
+        f"- **Tecnologias mais citadas**: {', '.join(top_tech) if top_tech else '—'}",
+        f"- **Menções internacionais**: ~{intl_hits}",
+        "\n> Observação: estimativas baseadas em busca por palavras-chave nos trechos indexados."
+    ]
+    return "\n".join(md)
+
+# ========= UI =========
+custom_css = r"""
+:root { --primary:#4a90e2; --bg-light:#f9f9f9; --txt-dark:#333; --radius:8px; --spacing:1rem; }
 body { background: var(--bg-light); color: var(--txt-dark); font-family: 'Helvetica Neue', sans-serif; }
 #chat-window { height: 65vh; overflow-y: auto; padding: var(--spacing); border: 1px solid #ddd; border-radius: var(--radius); }
-#input-area { display: flex; margin-top: var(--spacing); }
-#user-input { flex: 1; padding: 0.6rem; border: 1px solid #ccc; border-radius: var(--radius) 0 0 var(--radius); }
-#send-button { padding: 0 1rem; background: var(--primary); color: white; border: none; border-radius: 0 var(--radius) var(--radius) 0; cursor: pointer; }
 .sidebar { background: var(--bg-light); padding: var(--spacing); border-left: 1px solid #eee; }
-.sidebar h3 { margin-top: 0; }
 """
 
 with gr.Blocks(title=APP_TITLE, css=custom_css, theme=gr.themes.Base()) as demo:
@@ -159,37 +312,73 @@ with gr.Blocks(title=APP_TITLE, css=custom_css, theme=gr.themes.Base()) as demo:
     gr.Markdown(INTRO)
 
     with gr.Row():
-        # Main chat column
+        # Main chat
         with gr.Column(scale=3):
             chatbot_ui   = gr.Chatbot(type="tuples", elem_id="chat-window")
-            txt          = gr.Textbox(placeholder="Digite sua pergunta…", lines=2, elem_id="user-input")
-            btn          = gr.Button("Enviar", elem_id="send-button")
-            # Advanced settings sliders
-            temperature = gr.Slider(0, 1, value=0.6, label="Temperature")
-            top_p       = gr.Slider(0, 1, value=0.95, label="Top-p")
-            max_tokens  = gr.Slider(64, 2048, value=512, step=64, label="Max Tokens")
-
-            # Bind events
-            btn.click(chatbot, [txt, temperature, top_p, max_tokens], [chatbot_ui, txt])
+            txt          = gr.Textbox(placeholder="Digite sua pergunta…", lines=2)
+            btn_send     = gr.Button("Enviar", variant="primary")
+            btn_clear    = gr.Button("Limpar")
+
+            with gr.Accordion("Parâmetros avançados", open=False):
+                temperature = gr.Slider(0, 1, value=0.6, label="Temperature")
+                top_p       = gr.Slider(0, 1, value=0.95, label="Top-p")
+                max_tokens  = gr.Slider(64, 2048, value=512, step=64, label="Max Tokens")
+
+            btn_send.click(chatbot, [txt, temperature, top_p, max_tokens], [chatbot_ui, txt])
             txt.submit(chatbot, [txt, temperature, top_p, max_tokens], [chatbot_ui, txt])
-            gr.Button("Limpar").click(clear_history, [], [chatbot_ui, txt])
+            btn_clear.click(clear_history, [], [chatbot_ui, txt])
 
-        # Sidebar with PDF & suggestions
-        with gr.Column(scale=1, elem_classes="sidebar"):
+        # Sidebar
+        with gr.Column(scale=2, elem_classes="sidebar"):
             if Path(PDF_PATH).exists():
                 gr.Markdown(f"[📄 Baixar CV em PDF](/file={PDF_PATH})")
+
             gr.Markdown("### Sugestões de Perguntas")
             for q in SUGGESTION_QUESTIONS:
                 gr.Button(q).click(lambda suggestion=q: suggestion, outputs=[txt])
+
+            # (NEW) sugestões por tema
             gr.Markdown("---")
-            gr.Markdown("### Dicas de Exploração do PDF")
-            gr.Markdown("• Use palavras-chave como 'Process Mining' ou 'GIS' para ir direto à seção relevante.")
-            gr.Markdown("• Peça detalhes de projetos com financiamento público (CNPQ, QREN).")
+            gr.Markdown("### Sugestões por tema")
+            for theme, qs in SUGGESTIONS_THEMES.items():
+                with gr.Accordion(theme, open=False):
+                    for q in qs:
+                        gr.Button(q).click(lambda s=q: s, outputs=[txt])
 
-if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+            # (NEW) Mini-bios
+            gr.Markdown("---")
+            gr.Markdown("### Exportação rápida – Mini-bio (PDF)")
+            bio_style = gr.Dropdown(choices=list(MINI_BIO_STYLES.keys()), value="Corporativo", label="Formato")
+            btn_bio   = gr.Button("Gerar Mini-bio (PDF)")
+            bio_file  = gr.File(label="Mini-bio gerada")
+            bio_msg   = gr.Markdown()
+            btn_bio.click(generate_mini_bio, [bio_style, temperature, top_p, max_tokens], [bio_file, bio_msg])
 
+            # (NEW) Assistente de candidatura
+            gr.Markdown("---")
+            gr.Markdown("### Assistente de candidatura")
+            job_desc  = gr.Textbox(label="Cole a descrição da vaga", lines=8, placeholder="Cole aqui a JD…")
+            with gr.Row():
+                btn_cover = gr.Button("Gerar Carta (PDF)")
+                btn_match = gr.Button("Calcular Match Score")
+            cover_file = gr.File(label="Carta gerada")
+            cover_msg  = gr.Markdown()
+            match_out  = gr.Markdown()
 
+            btn_cover.click(generate_cover_letter, [job_desc, temperature, top_p, max_tokens], [cover_file, cover_msg])
+            btn_match.click(lambda jd: compute_match_score(jd), [job_desc], [match_out])
 
+            # (NEW) Métricas do CV
+            gr.Markdown("---")
+            gr.Markdown("### Métricas do CV")
+            btn_metrics = gr.Button("Recalcular métricas")
+            metrics_md  = gr.Markdown(value=extract_metrics())
+            btn_metrics.click(lambda: extract_metrics(), [], [metrics_md])
 
+            gr.Markdown("---")
+            gr.Markdown("### Dicas de Exploração do PDF")
+            gr.Markdown("• Use palavras-chave como 'Process Mining', 'GIS', 'Sentinel' para ir direto à seção relevante.")
+            gr.Markdown("• Peça detalhes de projetos financiados (CNPq, QREN, UE) e resultados mensuráveis.")
 
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)