Create app.py

app.py

@@ -0,0 +1,126 @@
+import openai, gradio as gr, json, plotly.graph_objects as go
+
+SYSTEM_PROMPT = """
+You are ZEN Multimodal Assistant created by ZEN AI Co.
+Choose the single best modality for each reply:
+
+1. Image – when an illustration, diagram, or photo-realistic scene answers best.  
+   Respond **only** with JSON:
+   {"type":"image","prompt":"<detailed prompt for DALL-E-3>"}
+
+2. Chart – when the user requests or clearly needs data-visualisation.  
+   Respond **only** with JSON:
+   {"type":"chart",
+    "title":"<chart title>",
+    "data":[
+        {"x":[...], "y":[...], "label":"<series name>"},
+        ...
+    ]}
+
+3. Text – for every other case.  
+   Respond with plain language (no JSON).
+
+Never wrap JSON in markdown fences and never add comments.
+"""
+
+# ---------- Helper functions ---------- #
+def build_messages(history, user_msg):
+    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+    for u, a in history:
+        messages.append({"role": "user", "content": u})
+        messages.append({"role": "assistant", "content": a})
+    messages.append({"role": "user", "content": user_msg})
+    return messages
+
+
+def multimodal_chat(api_key, user_msg, history):
+    if not api_key:
+        raise gr.Error("🔑  Please paste your OpenAI API key first.")
+    openai.api_key = api_key
+
+    history = history or []
+    messages = build_messages(history, user_msg)
+
+    response = openai.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=messages,
+        temperature=0.7,
+    )
+    assistant_content = response.choices[0].message.content.strip()
+
+    img_url, fig = None, None
+    try:                                                # JSON branch (image or chart)
+        parsed = json.loads(assistant_content)
+        if parsed.get("type") == "image":
+            dalle = openai.images.generate(
+                model="dall-e-3",
+                prompt=parsed.get("prompt", "high-quality illustration"),
+                n=1,
+                size="1024x1024",
+            )
+            img_url = dalle.data[0].url
+            history.append([user_msg, f"![generated image]({img_url})"])
+
+        elif parsed.get("type") == "chart":
+            fig = go.Figure()
+            for s in parsed["data"]:
+                fig.add_trace(
+                    go.Scatter(
+                        x=s["x"],
+                        y=s["y"],
+                        mode="lines+markers",
+                        name=s.get("label", ""),
+                    )
+                )
+            fig.update_layout(title=parsed.get("title", "Chart"))
+            history.append([user_msg, parsed.get("title", "Chart below")])
+
+        else:                                           # fallback to text
+            history.append([user_msg, assistant_content])
+
+    except (json.JSONDecodeError, KeyError, TypeError): # plain-text branch
+        history.append([user_msg, assistant_content])
+
+    return history, img_url, fig
+
+
+# ---------- Gradio UI ---------- #
+with gr.Blocks(title="ZEN Multimodal Assistant") as demo:
+    gr.Markdown(
+        """
+        # 🧠 ZEN Multimodal Assistant  
+        Paste your OpenAI API key once per session and chat freely –  
+        the assistant decides whether to answer with **text, an image, or an interactive chart**.
+        """
+    )
+
+    api_key   = gr.Textbox(label="OpenAI API Key", type="password", placeholder="sk-...")
+    chatbot   = gr.Chatbot(label="Conversation")
+    with gr.Row():
+        user_msg = gr.Textbox(placeholder="Ask me anything…", label="Your message", scale=4)
+        send_btn = gr.Button("Send", variant="primary")
+
+    img_out   = gr.Image(label="Generated image")
+    chart_out = gr.Plot(label="Interactive chart")
+
+    def respond(api_key, user_msg, chat_history):
+        chat_history, img_url, fig = multimodal_chat(api_key, user_msg, chat_history)
+
+        img_update  = gr.update(value=img_url) if img_url else gr.update(value=None)
+        fig_update  = gr.update(value=fig)     if fig     else gr.update(value=None)
+
+        return chat_history, img_update, fig_update
+
+    send_btn.click(
+        respond,
+        inputs=[api_key, user_msg, chatbot],
+        outputs=[chatbot, img_out, chart_out],
+    )
+    user_msg.submit(
+        respond,
+        inputs=[api_key, user_msg, chatbot],
+        outputs=[chatbot, img_out, chart_out],
+    )
+
+if __name__ == "__main__":
+    demo.queue(max_size=50).launch()