Create backend/app.py

backend/app.py

@@ -0,0 +1,153 @@
+# -----------------------------
+# Imports
+# -----------------------------
+import pandas as pd
+from google.colab import files
+import os
+import json
+from google import genai
+from google.genai import types
+
+# -----------------------------
+# Initialize Gemini client (global)
+# -----------------------------
+client = genai.Client(
+    api_key="AIzaSyB1jgGCuzg7ELPwNEEwaluQZoZhxhgLmAs"
+)
+
+# -----------------------------
+# Upload Excel file
+# -----------------------------
+uploaded = files.upload()
+file_name = list(uploaded.keys())[0]
+file_path = "/content/" + file_name
+df = pd.read_excel(file_name)
+
+# -----------------------------
+# Extract Metadata
+# -----------------------------
+def get_metadata(df):
+    return {
+        "columns": list(df.columns),
+        "dtypes": df.dtypes.apply(str).to_dict(),
+        "null_counts": df.isnull().sum().to_dict(),
+        "unique_counts": df.nunique().to_dict(),
+        "sample_rows": df.head(3).to_dict(orient="records")
+    }
+
+metadata = get_metadata(df)
+print("Metadata extracted:")
+print(metadata)
+
+# -----------------------------
+# Generate JSON summary and suggestions from metadata
+# -----------------------------
+def generate_metadata_analysis(metadata):
+    metadata_text = str(metadata)
+    model = "gemini-2.5-flash-lite"
+
+    contents = [
+        types.Content(
+            role="user",
+            parts=[types.Part.from_text(
+                text=f"Analyze the following structured data metadata:\n{metadata_text}"
+            )],
+        ),
+    ]
+
+    generate_content_config = types.GenerateContentConfig(
+        thinking_config=types.ThinkingConfig(thinking_budget=0),
+        response_mime_type="application/json",
+        system_instruction=[types.Part.from_text(text="""
+You are a structured data analysis AI.
+
+1️⃣ Summary:
+Provide a concise description of:
+- What kind of data this is
+- What it likely represents
+- Its domain or use-case
+Indicate assumptions if needed.
+
+2️⃣ Suggestions:
+Provide up to three actionable analyses and visualizations based on the metadata, specifying columns and insight type.
+
+Output must be strict JSON:
+{
+  "Summary": "<short summary>",
+  "Suggestion": ["<analysis #1>", "<analysis #2>", "<analysis #3>"]
+}
+""")],
+    )
+
+    output_text = ""
+    for chunk in client.models.generate_content_stream(
+        model=model,
+        contents=contents,
+        config=generate_content_config,
+    ):
+        output_text += chunk.text
+
+    return output_text
+
+# Call metadata analysis
+agent1summary_json = generate_metadata_analysis(metadata)
+agent1summary = json.loads(agent1summary_json)
+print("\nMetadata analysis JSON:")
+print(agent1summary)
+
+# -----------------------------
+# User selects one suggestion
+# -----------------------------
+print("\nSelect one of the following suggestions (type 1, 2, or 3):")
+for i, suggestion in enumerate(agent1summary["Suggestion"], 1):
+    print(f"{i}. {suggestion}")
+
+selected_index = int(input("Your selection: "))
+command = agent1summary["Suggestion"][selected_index - 1]
+print("\nSelected command:")
+print(command)
+
+# -----------------------------
+# Strict JSON output generator for visualization
+# -----------------------------
+MODEL = "gemini-2.5-flash-lite"
+
+system_prompt_text = f"""
+You are a Python assistant that MUST return output strictly in JSON format and NOTHING else.
+The top-level JSON MUST contain exactly three keys in this order: "type", "code", "explanation".
+Requirements:
+- "type": visualization type ("bar", "pie", "line", etc.)
+- "code": Python code as a string that prints numeric JSON to stdout. Use this for data access: df = pd.read_excel(r"{file_path}")
+- "explanation": one-sentence description
+"""
+
+def generate_visualization():
+    contents = [types.Content(role="user", parts=[types.Part.from_text(text=command)])]
+
+    generate_content_config = types.GenerateContentConfig(
+        thinking_config=types.ThinkingConfig(thinking_budget=0),
+        response_mime_type="application/json",
+        system_instruction=[types.Part.from_text(text=system_prompt_text)],
+    )
+
+    output = ""
+    for chunk in client.models.generate_content_stream(
+        model=MODEL,
+        contents=contents,
+        config=generate_content_config,
+    ):
+        output += chunk.text
+
+    return output
+
+# Call visualization generator
+agent2code = generate_visualization()
+print("\nStrict JSON for visualization:")
+print(agent2code)
+
+# -----------------------------
+# Execute generated visualization code
+# -----------------------------
+agent2code_json = json.loads(agent2code)
+code_to_run = agent2code_json.get("code", "")
+final_frontend_output = exec(code_to_run)