feat: multilingual support (ZH/ES/FR/JA) + remove all mock mode

- Add 4 new languages: Simplified Chinese, Spanish, French, Japanese
- Full native-language output: diagnosis, severity, actions, disclaimer
- Language selector upgraded from Radio to Dropdown (6 choices)
- Add EN·VI·ZH·ES·FR·JA badge in header
- Remove MOCK_MODE entirely — backend down = system offline (hard error)
- Strip all mock data pools, fallback logic, and MOCK_MODE env var
- Delete test_mock_pipeline.py
- Error card shown in UI when AMD Cloud is unreachable
- Status bar now shows "AMD Cloud · Offline" (red) instead of "Demo Mode"
- Update README: multilingual, remove mock mode docs, fix Gradio version

README.md

@@ -18,7 +18,7 @@ license: mit
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
 [![Powered by AMD](https://img.shields.io/badge/Powered%20by-AMD%20MI300X%20%2B%20ROCm-ED1C24)](https://www.amd.com/en/products/accelerators/instinct/mi300.html)
 
-MediVision is a bilingual (English / Vietnamese) multimodal AI assistant that analyzes skin wound and disease images combined with patient symptom descriptions. Inference is served by a **vLLM server running on AMD Developer Cloud** (AMD Instinct™ MI300X + ROCm), and the lightweight Gradio frontend on Hugging Face Spaces simply calls that API — no model weights are loaded in the Space itself.
+MediVision is a **multilingual** multimodal AI assistant that analyzes skin wound and disease images combined with patient symptom descriptions. It supports **English, Tiếng Việt, 中文, Español, Français, and 日本語** — every output (diagnosis, severity, actions, disclaimer) is delivered natively in the selected language. Inference is served by a **vLLM server running on AMD Developer Cloud** (AMD Instinct™ MI300X + ROCm), and the lightweight Gradio frontend on Hugging Face Spaces simply calls that API — no model weights are loaded in the Space itself.
 
 ---
 
@@ -52,14 +52,13 @@ on the AMD GPU server.
 ## Features
 
 - **Multimodal Analysis** — Combines skin image + freeform symptom text for richer diagnosis suggestions.
-- **Bilingual** — Full English and Vietnamese (Tiếng Việt) support.
+- **Multilingual** — Full support for 6 languages: English, Tiếng Việt, 中文 (Chinese), Español, Français, and 日本語 (Japanese). All output — diagnosis, severity, recommendations, and disclaimer — is rendered natively in the selected language.
 - **Structured Output** — Every analysis returns:
   - Diagnosis suggestion
   - Severity badge: `Low` · `Medium` · `High` · `Urgent`
   - Actionable recommended steps (clinical-grade language)
   - Confidence score with visual progress bar
 - **AMD MI300X Powered** — Inference via vLLM on AMD Instinct™ MI300X + ROCm.
-- **Graceful Mock Mode** — Falls back to realistic mock responses if the vLLM server is unreachable, so the demo always runs.
 - **HF Space Ready** — Minimal dependencies; no GPU required in the Space.
 
 ---
@@ -86,8 +85,8 @@ on the AMD GPU server.
 | Inference Runtime | **vLLM** (OpenAI-compatible API) |
 | Inference Hardware | **AMD Instinct™ MI300X** (192 GB HBM3) + ROCm |
 | Inference Host | AMD Developer Cloud |
-| Frontend | Gradio 4 (Hugging Face Space — CPU) |
-| Language Support | English / Tiếng Việt |
+| Frontend | Gradio 5.29 (Hugging Face Space — CPU) |
+| Language Support | English · Tiếng Việt · 中文 · Español · Français · 日本語 |
 
 ---
 
@@ -155,12 +154,6 @@ python app.py
 
 The app is available at `http://localhost:7860`.
 
-### 5. Force mock mode (no vLLM server required)
-
-```bash
-MOCK_MODE=true python app.py
-```
-
 ---
 
 ## Starting the vLLM Server (AMD Developer Cloud)
@@ -188,7 +181,7 @@ The server exposes an OpenAI-compatible API at `http://<VM_IP>:8000/v1`.
 | `VLLM_API_URL` | `http://localhost:8000` | Base URL of the vLLM server |
 | `MODEL_NAME` | `Qwen/Qwen2.5-VL-7B-Instruct` | Model ID served by vLLM |
 | `VLLM_API_KEY` | `not-required` | API key (if vLLM auth is enabled) |
-| `MOCK_MODE` | `false` | Force mock mode (skip vLLM calls) |
+
 | `MAX_NEW_TOKENS` | `512` | Max tokens to generate |
 | `TEMPERATURE` | `0.2` | Sampling temperature |
 

app.py

@@ -1,7 +1,6 @@
 import gradio as gr
 from src.inference import MediVisionPipeline
 from src.model_loader import check_connection
-import src.config as config
 
 # ---------------------------------------------------------------------------
 # Pipeline singleton
@@ -25,7 +24,7 @@ def get_backend_status_html() -> str:
     if connected:
         dot, label, color = "#22c55e", "AMD Cloud · Live", "#86efac"
     else:
-        dot, label, color = "#f97316", "Demo Mode", "#fdba74"
+        dot, label, color = "#ef4444", "AMD Cloud · Offline", "#fca5a5"
     return (
         f"<div style='text-align:center; margin-bottom:4px;'>"
         f"<span style='font-size:0.75rem; color:{color}; font-family:monospace;'>"
@@ -39,14 +38,36 @@ def get_backend_status_html() -> str:
 # ---------------------------------------------------------------------------
 
 _SEVERITY_COLOR = {
-    "Low":       ("#22c55e", "#dcfce7"),
-    "Medium":    ("#eab308", "#fef9c3"),
-    "High":      ("#f97316", "#ffedd5"),
-    "Urgent":    ("#ef4444", "#fee2e2"),
-    "Thấp":      ("#22c55e", "#dcfce7"),
+    # English
+    "Low":        ("#22c55e", "#dcfce7"),
+    "Medium":     ("#eab308", "#fef9c3"),
+    "High":       ("#f97316", "#ffedd5"),
+    "Urgent":     ("#ef4444", "#fee2e2"),
+    # Vietnamese
+    "Thấp":       ("#22c55e", "#dcfce7"),
     "Trung bình": ("#eab308", "#fef9c3"),
-    "Cao":       ("#f97316", "#ffedd5"),
-    "Khẩn cấp":  ("#ef4444", "#fee2e2"),
+    "Cao":        ("#f97316", "#ffedd5"),
+    "Khẩn cấp":   ("#ef4444", "#fee2e2"),
+    # Chinese
+    "低":          ("#22c55e", "#dcfce7"),
+    "中":          ("#eab308", "#fef9c3"),
+    "高":          ("#f97316", "#ffedd5"),
+    "紧急":        ("#ef4444", "#fee2e2"),
+    # Spanish
+    "Baja":       ("#22c55e", "#dcfce7"),
+    "Media":      ("#eab308", "#fef9c3"),
+    "Alta":       ("#f97316", "#ffedd5"),
+    "Urgente":    ("#ef4444", "#fee2e2"),
+    # French
+    "Faible":     ("#22c55e", "#dcfce7"),
+    "Modérée":    ("#eab308", "#fef9c3"),
+    "Élevée":     ("#f97316", "#ffedd5"),
+    "Urgente":    ("#ef4444", "#fee2e2"),
+    # Japanese
+    "軽度":        ("#22c55e", "#dcfce7"),
+    "中等度":      ("#eab308", "#fef9c3"),
+    "重度":        ("#f97316", "#ffedd5"),
+    "緊急":        ("#ef4444", "#fee2e2"),
 }
 
 
@@ -81,42 +102,23 @@ def _build_result_html(result: dict, lang: str) -> str:
     actions = result.get("recommended_actions", [])
     score   = result.get("confidence_score", 0)
 
-    if lang == "vn":
-        diag_label     = "Gợi ý chẩn đoán"
-        severity_label = "Mức độ nghiêm trọng"
-        actions_label  = "Khuyến nghị"
-        disclaimer = (
-            "Đây là trợ lý AI, không phải bác sĩ. "
-            "Hãy tham khảo chuyên gia y tế cho các tình trạng nghiêm trọng."
-        )
-    else:
-        diag_label     = "Diagnosis Suggestion"
-        severity_label = "Severity"
-        actions_label  = "Recommended Actions"
-        disclaimer = (
-            "This is an AI assistant, not a licensed physician. "
-            "Always consult a healthcare professional for serious conditions."
-        )
+    t = _I18N.get(lang, _I18N["en"])
+    diag_label     = t["diag_label"]
+    severity_label = t["severity_label"]
+    actions_label  = t["actions_label"]
+    disclaimer     = t["disclaimer"]
 
     actions_html = "".join(
         f"<li style='margin:5px 0; color:#d1d5db;'>{a}</li>"
         for a in actions
     ) if actions else "<li style='color:#6b7280;'>—</li>"
 
-    if config.MOCK_MODE:
-        backend_tag = (
-            "<span style='font-size:0.7rem; background:#431407; color:#fdba74; "
-            "padding:2px 8px; border-radius:4px; margin-left:8px; "
-            "border:1px solid #c2410c;'>Demo Mode</span>"
-        )
-        backend_info = "Demo Mode · Representative Response"
-    else:
-        backend_tag = (
-            "<span style='font-size:0.7rem; background:#052e16; color:#86efac; "
-            "padding:2px 8px; border-radius:4px; margin-left:8px; "
-            "border:1px solid #16a34a;'>AMD Cloud</span>"
-        )
-        backend_info = "AMD MI300X · ROCm · Qwen2.5-VL-7B"
+    backend_tag = (
+        "<span style='font-size:0.7rem; background:#052e16; color:#86efac; "
+        "padding:2px 8px; border-radius:4px; margin-left:8px; "
+        "border:1px solid #16a34a;'>AMD Cloud</span>"
+    )
+    backend_info = "AMD MI300X · ROCm · Qwen2.5-VL-7B"
 
     return f"""
 <div style='background:#111827; border:1px solid #ED1C24; border-radius:12px;
@@ -175,22 +177,106 @@ def _build_result_html(result: dict, lang: str) -> str:
 # Main prediction function
 # ---------------------------------------------------------------------------
 
+_LANG_MAP = {
+    "English":    "en",
+    "Tiếng Việt": "vn",
+    "中文":        "zh",
+    "Español":    "es",
+    "Français":   "fr",
+    "日本語":      "ja",
+}
+
+_I18N = {
+    "en": {
+        "diag_label":     "Diagnosis Suggestion",
+        "severity_label": "Severity",
+        "actions_label":  "Recommended Actions",
+        "disclaimer": (
+            "This is an AI assistant, not a licensed physician. "
+            "Always consult a healthcare professional for serious conditions."
+        ),
+        "placeholder": "Please upload an image or enter symptoms.",
+    },
+    "vn": {
+        "diag_label":     "Gợi ý chẩn đoán",
+        "severity_label": "Mức độ nghiêm trọng",
+        "actions_label":  "Khuyến nghị",
+        "disclaimer": (
+            "Đây là trợ lý AI, không phải bác sĩ. "
+            "Hãy tham khảo chuyên gia y tế cho các tình trạng nghiêm trọng."
+        ),
+        "placeholder": "Vui lòng tải lên hình ảnh hoặc nhập triệu chứng.",
+    },
+    "zh": {
+        "diag_label":     "诊断建议",
+        "severity_label": "严重程度",
+        "actions_label":  "推荐措施",
+        "disclaimer": (
+            "本工具为AI助手，不能替代执业医师。"
+            "如有严重病情，请务必咨询专业医疗人员。"
+        ),
+        "placeholder": "请上传图片或输入症状描述。",
+    },
+    "es": {
+        "diag_label":     "Sugerencia de diagnóstico",
+        "severity_label": "Severidad",
+        "actions_label":  "Acciones recomendadas",
+        "disclaimer": (
+            "Este es un asistente de IA, no un médico autorizado. "
+            "Consulte siempre a un profesional de la salud para condiciones graves."
+        ),
+        "placeholder": "Por favor, suba una imagen o describa sus síntomas.",
+    },
+    "fr": {
+        "diag_label":     "Suggestion de diagnostic",
+        "severity_label": "Sévérité",
+        "actions_label":  "Actions recommandées",
+        "disclaimer": (
+            "Ceci est un assistant IA, pas un médecin agréé. "
+            "Consultez toujours un professionnel de santé pour les situations graves."
+        ),
+        "placeholder": "Veuillez télécharger une image ou décrire vos symptômes.",
+    },
+    "ja": {
+        "diag_label":     "診断提案",
+        "severity_label": "重症度",
+        "actions_label":  "推奨アクション",
+        "disclaimer": (
+            "これはAIアシスタントであり、有資格の医師ではありません。"
+            "深刻な症状については必ず医療専門家に相談してください。"
+        ),
+        "placeholder": "画像をアップロードするか、症状を入力してください。",
+    },
+}
+
+
 def predict(image, symptoms: str, lang_choice: str):
-    lang = "vn" if lang_choice == "Tiếng Việt" else "en"
+    lang = _LANG_MAP.get(lang_choice, "en")
 
     if not image and not symptoms.strip():
-        placeholder = (
-            "Please upload an image or enter symptoms."
-            if lang == "en"
-            else "Vui lòng tải lên hình ảnh hoặc nhập triệu chứng."
-        )
+        placeholder = _I18N.get(lang, _I18N["en"])["placeholder"]
         return (
             f"<p style='color:#9ca3af; text-align:center;'>{placeholder}</p>",
             get_backend_status_html(),
         )
 
-    result = get_pipeline().process(image, symptoms.strip(), lang=lang)
-    return _build_result_html(result, lang), get_backend_status_html()
+    try:
+        result = get_pipeline().process(image, symptoms.strip(), lang=lang)
+        return _build_result_html(result, lang), get_backend_status_html()
+    except Exception as exc:
+        error_html = (
+            "<div style='background:#111827; border:1px solid #ef4444; border-radius:12px;"
+            "padding:24px; font-family:Arial,sans-serif; text-align:center;'>"
+            "<div style='font-size:1.5rem; margin-bottom:12px;'>⚠️</div>"
+            "<div style='font-size:1rem; font-weight:700; color:#ef4444; margin-bottom:8px;'>"
+            "Backend Unavailable / Hệ thống không khả dụng</div>"
+            "<div style='font-size:0.85rem; color:#9ca3af; margin-bottom:16px;'>"
+            "AMD Cloud backend is unreachable. Please try again later.</div>"
+            f"<div style='font-size:0.75rem; color:#6b7280; font-family:monospace;"
+            f"background:#1f2937; padding:8px 12px; border-radius:6px;'>{exc}</div>"
+            "</div>"
+        )
+        return error_html, get_backend_status_html()
 
 
 # ---------------------------------------------------------------------------
@@ -257,7 +343,7 @@ HEADER_HTML = """
     <span style='color:#ED1C24;'>Medi</span><span style='color:#f9fafb;'>Vision</span>
   </div>
   <div style='color:#9ca3af; font-size:0.9rem; margin-top:4px;'>
-    Multimodal Medical Imaging AI Agent
+    Multilingual Multimodal Medical Imaging AI Agent
   </div>
   <div style='margin-top:10px; display:inline-flex; gap:8px; flex-wrap:wrap;
               justify-content:center;'>
@@ -269,6 +355,10 @@ HEADER_HTML = """
                  padding:3px 12px; border-radius:999px; border:1px solid #374151;'>
       ROCm · Qwen2.5-VL-7B
     </span>
+    <span style='background:#1f2937; color:#9ca3af; font-size:0.72rem; font-weight:600;
+                 padding:3px 12px; border-radius:999px; border:1px solid #374151;'>
+      EN · VI · ZH · ES · FR · JA
+    </span>
     <span style='background:#1f2937; color:#9ca3af; font-size:0.72rem; font-weight:600;
                  padding:3px 12px; border-radius:999px; border:1px solid #374151;'>
       AMD Developer Hackathon 2026
@@ -312,8 +402,8 @@ with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="MediVision — AMD MI300X
                 ),
                 lines=4,
             )
-            lang_radio = gr.Radio(
-                choices=["English", "Tiếng Việt"],
+            lang_radio = gr.Dropdown(
+                choices=["English", "Tiếng Việt", "中文", "Español", "Français", "日本語"],
                 value="English",
                 label="Language / Ngôn ngữ",
             )
@@ -328,7 +418,10 @@ with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="MediVision — AMD MI300X
                     [None, "I have a red, itchy rash on my forearm for 3 days. It burns slightly.", "English"],
                     [None, "Small wound on my hand after a cut, slightly swollen with some redness.", "English"],
                     [None, "Vết thương nhỏ ở bàn tay, hơi sưng và có dấu hiệu đỏ xung quanh.", "Tiếng Việt"],
-                    [None, "Phát ban đỏ trên cánh tay, ngứa nhiều, đã 2 ngày.", "Tiếng Việt"],
+                    [None, "手臂上出现红色瘙痒皮疹，已持续3天，略有灼热感。", "中文"],
+                    [None, "Tengo una erupción roja y con picazón en el antebrazo desde hace 3 días.", "Español"],
+                    [None, "J'ai une éruption rouge et prurigineuse sur l'avant-bras depuis 3 jours.", "Français"],
+                    [None, "3日前から前腕に赤くてかゆい発疹があり、少し灼熱感があります。", "日本語"],
                 ],
                 inputs=[input_img, symptoms_txt, lang_radio],
                 label="Quick Examples / Ví dụ nhanh",

src/agent.py

@@ -1,207 +1,21 @@
-import random
+import json
 import re
-import src.config as config
 from src.model_loader import generate_response
 
-# ---------------------------------------------------------------------------
-# Mock data pools — realistic enough for a hackathon demo
-# ---------------------------------------------------------------------------
+# Supported language codes: en, vn, zh, es, fr, ja
 
-_MOCK_EN = [
-    {
-        "diagnosis": "Contact Dermatitis",
-        "severity": "Low",
-        "confidence_score": 82,
-        "recommended_actions": [
-            "Avoid contact with suspected allergen or irritant.",
-            "Apply a mild hydrocortisone cream (1%) to the affected area twice daily.",
-            "Use fragrance-free moisturiser after bathing to restore the skin barrier.",
-            "Take an over-the-counter antihistamine (e.g., cetirizine) if itching is severe.",
-            "Monitor for spreading redness or signs of secondary infection; see a doctor if worsening.",
-        ],
-    },
-    {
-        "diagnosis": "Mild Abrasion / Superficial Wound",
-        "severity": "Low",
-        "confidence_score": 88,
-        "recommended_actions": [
-            "Gently clean the wound with sterile saline or clean running water for 5–10 minutes.",
-            "Apply a thin layer of antibiotic ointment (e.g., bacitracin) to prevent infection.",
-            "Cover with a non-stick sterile dressing; change daily or when soiled.",
-            "Watch for signs of infection: increased redness, warmth, purulent discharge, or fever.",
-            "Tetanus prophylaxis: verify vaccination is up to date (within 5 years for dirty wounds).",
-        ],
-    },
-    {
-        "diagnosis": "Atopic Eczema (Eczema Flare)",
-        "severity": "Medium",
-        "confidence_score": 75,
-        "recommended_actions": [
-            "Apply a prescription-strength topical corticosteroid (e.g., triamcinolone 0.1%) to inflamed areas.",
-            "Use an emollient at least twice daily on the entire body, not just affected skin.",
-            "Identify and eliminate triggers: dust mites, pet dander, harsh soaps, or stress.",
-            "Wear loose-fitting, breathable cotton clothing to minimise irritation.",
-            "Consult a dermatologist if the flare does not improve within 1–2 weeks.",
-        ],
-    },
-    {
-        "diagnosis": "Partial-Thickness Burn (Second-Degree)",
-        "severity": "High",
-        "confidence_score": 79,
-        "recommended_actions": [
-            "Cool the burn immediately under cool (not ice-cold) running water for at least 20 minutes.",
-            "Do NOT apply butter, toothpaste, or home remedies — these increase infection risk.",
-            "Cover loosely with a clean non-fluffy material (e.g., cling film or sterile dressing).",
-            "Seek prompt medical evaluation; burns larger than a palm or on the face/hands require ER care.",
-            "Pain management: ibuprofen or paracetamol as directed.",
-        ],
-    },
-    {
-        "diagnosis": "Suspected Cellulitis",
-        "severity": "Urgent",
-        "confidence_score": 71,
-        "recommended_actions": [
-            "Seek immediate medical attention — cellulitis can spread rapidly and become systemic.",
-            "Do not massage or heat the affected area.",
-            "Elevate the limb above heart level to reduce oedema.",
-            "A course of oral antibiotics (e.g., cephalexin or amoxicillin-clavulanate) is typically required.",
-            "Return to ER if red streaking, fever > 38.5 °C, or rapid area expansion occurs.",
-        ],
-    },
-    {
-        "diagnosis": "Tinea Corporis (Ringworm)",
-        "severity": "Low",
-        "confidence_score": 85,
-        "recommended_actions": [
-            "Apply an antifungal cream (e.g., clotrimazole 1% or terbinafine) twice daily for 2–4 weeks.",
-            "Keep the area clean and dry; fungi thrive in moist environments.",
-            "Avoid sharing towels, clothing, or bedding during active infection.",
-            "Wash clothing and bed linen at 60 °C to eliminate spores.",
-            "See a GP if there is no improvement after 4 weeks — oral antifungal may be needed.",
-        ],
-    },
-    {
-        "diagnosis": "Psoriasis Plaque",
-        "severity": "Medium",
-        "confidence_score": 73,
-        "recommended_actions": [
-            "Moisturise heavily with thick emollients (e.g., petroleum jelly) after bathing.",
-            "A topical vitamin D analogue (e.g., calcipotriol) or mild steroid may be prescribed.",
-            "Avoid known triggers: stress, alcohol, smoking, and certain medications (e.g., beta-blockers).",
-            "Phototherapy (UVB) is effective for extensive plaques — discuss with a dermatologist.",
-            "Biologic therapy may be appropriate for moderate-to-severe disease unresponsive to topicals.",
-        ],
-    },
-]
-
-_MOCK_VN = [
-    {
-        "diagnosis": "Viêm da tiếp xúc",
-        "severity": "Thấp",
-        "confidence_score": 82,
-        "recommended_actions": [
-            "Tránh tiếp xúc với chất gây dị ứng hoặc kích ứng nghi ngờ.",
-            "Bôi kem hydrocortisone nhẹ (1%) lên vùng bị ảnh hưởng hai lần mỗi ngày.",
-            "Dùng kem dưỡng ẩm không hương liệu sau khi tắm để phục hồi hàng rào da.",
-            "Uống thuốc kháng histamine (ví dụ: cetirizine) nếu ngứa nghiêm trọng.",
-            "Theo dõi xem có lan rộng đỏ hoặc dấu hiệu nhiễm trùng thứ phát không; gặp bác sĩ nếu nặng hơn.",
-        ],
-    },
-    {
-        "diagnosis": "Trầy xước nhẹ / Vết thương nông",
-        "severity": "Thấp",
-        "confidence_score": 88,
-        "recommended_actions": [
-            "Nhẹ nhàng làm sạch vết thương bằng nước muối sinh lý hoặc nước sạch trong 5–10 phút.",
-            "Bôi một lớp mỏng thuốc mỡ kháng sinh (ví dụ: bacitracin) để ngăn ngừa nhiễm trùng.",
-            "Băng bó bằng gạc vô khuẩn không dính; thay hàng ngày hoặc khi bẩn.",
-            "Theo dõi dấu hiệu nhiễm trùng: đỏ tăng, nóng, mủ chảy ra hoặc sốt.",
-            "Phòng uốn ván: kiểm tra lịch tiêm chủng có còn hiệu lực (trong vòng 5 năm với vết thương bẩn).",
-        ],
-    },
-    {
-        "diagnosis": "Chàm dị ứng (Đợt bùng phát)",
-        "severity": "Trung bình",
-        "confidence_score": 75,
-        "recommended_actions": [
-            "Bôi corticosteroid tại chỗ theo toa (ví dụ: triamcinolone 0.1%) lên vùng viêm.",
-            "Dùng thuốc dưỡng ẩm ít nhất hai lần mỗi ngày trên toàn cơ thể, không chỉ vùng bị ảnh hưởng.",
-            "Xác định và loại bỏ các tác nhân kích thích: bụi, lông thú cưng, xà phòng mạnh hoặc căng thẳng.",
-            "Mặc quần áo cotton rộng rãi, thoáng khí để giảm kích ứng.",
-            "Tham khảo bác sĩ da liễu nếu đợt bùng phát không cải thiện trong vòng 1–2 tuần.",
-        ],
-    },
-    {
-        "diagnosis": "Bỏng độ hai (Bỏng lớp bì)",
-        "severity": "Cao",
-        "confidence_score": 79,
-        "recommended_actions": [
-            "Làm mát vết bỏng ngay bằng nước mát (không phải nước đá) trong ít nhất 20 phút.",
-            "KHÔNG bôi bơ, kem đánh răng hoặc các biện pháp dân gian — sẽ tăng nguy cơ nhiễm trùng.",
-            "Che phủ nhẹ nhàng bằng vải sạch không xơ (ví dụ: màng bọc thực phẩm hoặc gạc vô khuẩn).",
-            "Tìm kiếm đánh giá y tế ngay; bỏng lớn hơn lòng bàn tay hoặc ở mặt/tay cần đến cấp cứu.",
-            "Giảm đau: ibuprofen hoặc paracetamol theo chỉ định.",
-        ],
-    },
-    {
-        "diagnosis": "Nghi ngờ viêm mô tế bào (Cellulitis)",
-        "severity": "Khẩn cấp",
-        "confidence_score": 71,
-        "recommended_actions": [
-            "Tìm kiếm sự chú ý y tế ngay lập tức — viêm mô tế bào có thể lan rộng nhanh chóng.",
-            "Không xoa bóp hoặc chườm nóng vùng bị ảnh hưởng.",
-            "Nâng cao chi trên mức tim để giảm phù nề.",
-            "Thường cần một đợt kháng sinh uống (ví dụ: cephalexin hoặc amoxicillin-clavulanate).",
-            "Quay lại cấp cứu nếu có vết đỏ lan rộng, sốt > 38.5 °C hoặc vùng bị ảnh hưởng mở rộng nhanh.",
-        ],
-    },
-    {
-        "diagnosis": "Nấm da (Ringworm / Tinea corporis)",
-        "severity": "Thấp",
-        "confidence_score": 85,
-        "recommended_actions": [
-            "Bôi kem chống nấm (ví dụ: clotrimazole 1% hoặc terbinafine) hai lần mỗi ngày trong 2–4 tuần.",
-            "Giữ vùng da sạch và khô; nấm phát triển mạnh trong môi trường ẩm ướt.",
-            "Tránh dùng chung khăn, quần áo hoặc chăn ga trong thời gian nhiễm trùng.",
-            "Giặt quần áo và ga trải giường ở nhiệt độ 60 °C để tiêu diệt bào tử.",
-            "Gặp bác sĩ nếu không cải thiện sau 4 tuần — có thể cần thuốc chống nấm uống.",
-        ],
-    },
-    {
-        "diagnosis": "Mảng vảy nến (Psoriasis Plaque)",
-        "severity": "Trung bình",
-        "confidence_score": 73,
-        "recommended_actions": [
-            "Dưỡng ẩm nhiều bằng các chất nhũ hóa dày (ví dụ: vaseline) sau khi tắm.",
-            "Có thể kê đơn thuốc tương tự vitamin D tại chỗ (ví dụ: calcipotriol) hoặc steroid nhẹ.",
-            "Tránh các yếu tố kích thích đã biết: căng thẳng, rượu, hút thuốc và một số thuốc.",
-            "Quang trị liệu (UVB) có hiệu quả với các mảng rộng — thảo luận với bác sĩ da liễu.",
-            "Liệu pháp sinh học có thể phù hợp với bệnh từ vừa đến nặng không đáp ứng với thuốc tại chỗ.",
-        ],
-    },
-]
-
-_SEVERITY_ORDER = {"Low": 0, "Thấp": 0, "Medium": 1, "Trung bình": 1,
-                   "High": 2, "Cao": 2, "Urgent": 3, "Khẩn cấp": 3}
-
-
-def _add_variance(base_score: int) -> int:
-    """Vary the confidence score ±5 points within [65, 95]."""
-    return max(65, min(95, base_score + random.randint(-5, 5)))
-
-
-def _mock_response(lang: str) -> dict:
-    pool = _MOCK_VN if lang == "vn" else _MOCK_EN
-    entry = random.choice(pool).copy()
-    entry["confidence_score"] = _add_variance(entry["confidence_score"])
-    return entry
+_LANG_INSTRUCTIONS = {
+    "en": "Respond in English.",
+    "vn": "Respond in Vietnamese (Tiếng Việt).",
+    "zh": "Respond in Simplified Chinese (简体中文).",
+    "es": "Respond in Spanish (Español).",
+    "fr": "Respond in French (Français).",
+    "ja": "Respond in Japanese (日本語).",
+}
 
 
 def _build_prompt(image_path: str | None, text_description: str, lang: str) -> str:
-    lang_instruction = (
-        "Respond in Vietnamese." if lang == "vn"
-        else "Respond in English."
-    )
+    lang_instruction = _LANG_INSTRUCTIONS.get(lang, _LANG_INSTRUCTIONS["en"])
     has_image = bool(image_path)
     return (
         "You are MediVision, a professional dermatology and wound-care assistant.\n"
@@ -216,14 +30,7 @@ def _build_prompt(image_path: str | None, text_description: str, lang: str) -> s
     )
 
 
-def _parse_real_response(raw: str, lang: str) -> dict:
-    """
-    Try to extract JSON from the model's raw output.
-    Falls back to a structured mock on parse failure.
-    """
-    import json
-
-    # Find the first JSON object in the output
+def _parse_response(raw: str) -> dict:
     match = re.search(r'\{.*\}', raw, re.DOTALL)
     if match:
         try:
@@ -236,11 +43,7 @@ def _parse_real_response(raw: str, lang: str) -> dict:
             }
         except (json.JSONDecodeError, ValueError):
             pass
-
-    # Couldn't parse — use mock with a note
-    result = _mock_response(lang)
-    result["diagnosis"] = f"[Parse error — showing representative result] {result['diagnosis']}"
-    return result
+    raise ValueError(f"Could not parse model response as JSON: {raw[:200]}")
 
 
 def analyze_image_and_text(
@@ -249,26 +52,11 @@ def analyze_image_and_text(
     language: str = "en",
 ) -> dict:
     """
-    Main analysis entry point.
-
-    Returns a dict:
-        {
-            "diagnosis": str,
-            "severity": str,          # Low | Medium | High | Urgent
-            "recommended_actions": list[str],
-            "confidence_score": int,   # 0-100
-        }
+    Run analysis via AMD Cloud backend.
+    Raises RuntimeError if the backend is unreachable.
+    Raises ValueError if the model response cannot be parsed.
     """
     lang = language.lower()
-
-    if config.MOCK_MODE:
-        return _mock_response(lang)
-
     prompt = _build_prompt(image_path, text_description, lang)
     raw = generate_response(prompt, image_path=image_path)
-
-    if raw is None:
-        # model_loader set MOCK_MODE=True due to load failure
-        return _mock_response(lang)
-
-    return _parse_real_response(raw, lang)
+    return _parse_response(raw)

src/config.py

@@ -5,9 +5,6 @@ VLLM_API_URL = os.environ.get("VLLM_API_URL", "http://localhost:8000")
 
 MODEL_NAME = os.environ.get("MODEL_NAME", "Qwen/Qwen2.5-VL-7B-Instruct")
 
-# Set to True to force mock mode (skip vLLM calls entirely)
-MOCK_MODE = os.environ.get("MOCK_MODE", "false").lower() == "true"
-
 # Generation settings
 MAX_NEW_TOKENS = int(os.environ.get("MAX_NEW_TOKENS", "512"))
 TEMPERATURE = float(os.environ.get("TEMPERATURE", "0.2"))

src/inference.py

@@ -5,23 +5,11 @@ class MediVisionPipeline:
     def process(self, image_path, symptoms: str, lang: str = "en") -> dict:
         """
         Run the full analysis pipeline.
+        Raises RuntimeError if the AMD Cloud backend is unreachable.
 
         Returns:
             dict with keys: diagnosis, severity, recommended_actions, confidence_score
         """
-        if not image_path and not symptoms.strip():
-            placeholder = (
-                "Please upload an image or describe your symptoms."
-                if lang == "en"
-                else "Vui lòng tải lên hình ảnh hoặc mô tả triệu chứng của bạn."
-            )
-            return {
-                "diagnosis": placeholder,
-                "severity": "Low",
-                "recommended_actions": [],
-                "confidence_score": 0,
-            }
-
         return analyze_image_and_text(
             image_path=image_path,
             text_description=symptoms,

src/model_loader.py

@@ -36,10 +36,6 @@ def check_connection() -> tuple[bool, str]:
     Ping the vLLM server's /v1/models endpoint.
     Returns (is_connected: bool, status_message: str).
     """
-    if config.MOCK_MODE:
-        print("[Connection] MOCK_MODE=true — skipping connection check.")
-        return False, "Mock mode enabled"
-
     import requests as req
 
     url = f"{config.VLLM_API_URL}/v1/models"
@@ -58,21 +54,18 @@ def check_connection() -> tuple[bool, str]:
         print(f"[Connection] FAILED — ConnectionError: {exc}")
         return False, f"ConnectionError: {exc}"
     except req.exceptions.Timeout:
-        print(f"[Connection] FAILED — Timeout after 5s (server may be down or IP wrong)")
+        print(f"[Connection] FAILED — Timeout after 5s")
         return False, "Timeout (5s)"
     except Exception as exc:
-        print(f"[Connection] FAILED — Unexpected error: {type(exc).__name__}: {exc}")
+        print(f"[Connection] FAILED — {type(exc).__name__}: {exc}")
         return False, f"{type(exc).__name__}: {exc}"
 
 
-def generate_response(prompt: str, image_path: str = None) -> str | None:
+def generate_response(prompt: str, image_path: str = None) -> str:
     """
     Send a request to the vLLM endpoint and return the model's text output.
-    Returns None on failure so callers fall back to mock logic.
+    Raises RuntimeError if the backend is unreachable or returns an error.
     """
-    if config.MOCK_MODE:
-        return None
-
     try:
         client = _get_client()
 
@@ -84,9 +77,7 @@ def generate_response(prompt: str, image_path: str = None) -> str | None:
                     "content": [
                         {
                             "type": "image_url",
-                            "image_url": {
-                                "url": f"data:{mime};base64,{b64}",
-                            },
+                            "image_url": {"url": f"data:{mime};base64,{b64}"},
                         },
                         {"type": "text", "text": prompt},
                     ],
@@ -104,6 +95,4 @@ def generate_response(prompt: str, image_path: str = None) -> str | None:
         return response.choices[0].message.content
 
     except Exception as exc:
-        print(f"[ModelLoader] vLLM call failed ({exc}). Falling back to mock mode.")
-        config.MOCK_MODE = True
-        return None
+        raise RuntimeError(f"AMD Cloud backend unreachable: {exc}") from exc

test_mock_pipeline.py

@@ -1,20 +0,0 @@
-from src.inference import MediVisionPipeline
-import os
-
-def test_pipeline():
-    print("Initializing Pipeline...")
-    pipeline = MediVisionPipeline()
-    print("Pipeline Initialized.")
-    
-    image_path = None # Testing without image first
-    symptoms = "I have a red rash on my arm that is itchy."
-    
-    print(f"Testing with symptoms: {symptoms}")
-    result = pipeline.process(image_path, symptoms, lang="en")
-    print(f"Result (EN): {result}")
-    
-    result_vn = pipeline.process(image_path, symptoms, lang="vn")
-    print(f"Result (VN): {result_vn}")
-
-if __name__ == "__main__":
-    test_pipeline()