Keyword inspector card added.Cleaned up Search term code

app.py

@@ -12,6 +12,7 @@ print("IMPORT 3 OK", flush=True)
 from app.ads1.ads_analyst import run_ads_analyst_card
 print("IMPORT 4 OK", flush=True)
 from app.ads1.search_term_optimizer import run_search_term_optimizer
+from app.ads1.keyword_inspector import run_keyword_inspector
 
 # ==================================================
 # ROMER / ADVISOR DASHBOARD THEME
@@ -978,6 +979,22 @@ def run_search_term_optimizer_card(state):
     except Exception as e:
         return f"Search term optimization failed: {e}"
 
+@spaces.GPU(duration=120)
+def run_keyword_inspector_card(state):
+    try:
+        if not state:
+            return "Select a campaign first."
+
+        dfs = state.get("full_dfs")
+        campaign_name = state.get("campaign_name")
+
+        if dfs is None or campaign_name is None:
+            return "Campaign state is not properly initialized."
+
+        return run_keyword_inspector(dfs, campaign_name=campaign_name)
+
+    except Exception as e:
+        return f"Search term optimization failed: {e}"
 
 # ==================================================
 # UI
@@ -1021,7 +1038,10 @@ with gr.Blocks(fill_height=True, fill_width=True, css=CSS) as demo:
                         elem_classes=["ai-button-card"],
                     )
                     gr.HTML(ai_card("Budget Optimizer", "Where to adjust spend?"))
-                    gr.HTML(ai_card("Keyword Inspector", "Winning versus wasting keywords."))
+                    keyword_inspector_card = gr.Button(
+                        value="Keyword Inspector\n <hr> Winning versus wasting keywords..",
+                        elem_classes=["ai-button-card"],
+                    )
 
                 with gr.Row(elem_classes=["ai-row"]):
                     search_term_card = gr.Button(
@@ -1057,6 +1077,12 @@ with gr.Blocks(fill_height=True, fill_width=True, css=CSS) as demo:
         outputs=ads_output,
     )
 
+    keyword_inspector_card.click(
+        fn=run_keyword_inspector_card,
+        inputs=campaign_state,
+        outputs=ads_output,
+    )
+
     demo.load(
         fn=initial_data_load,
         outputs=[full_state, campaign_picker, hero_html, kpi_html],

app/ads1/keyword_inspector.py

@@ -0,0 +1,93 @@
+import json
+import pandas as pd
+
+from app.recs.generate import generate_explanation, is_bad_llm_output
+
+
+# -------------------------
+# Feature engineering only
+# -------------------------
+def build_keyword_features(df: pd.DataFrame) -> pd.DataFrame:
+    df = df.copy()
+
+    df["cost"] = df["cost"].fillna(0)
+    df["clicks"] = df["clicks"].fillna(0)
+    df["impressions"] = df["impressions"].fillna(0)
+    df["conversions"] = df.get("conversions", 0).fillna(0)
+
+    df["ctr"] = (df["clicks"] / df["impressions"].replace(0, 1)) * 100
+    df["cpa"] = df["cost"] / df["conversions"].replace(0, 1)
+
+    return df
+
+
+# -------------------------
+# Prompt (simplified + stronger reasoning)
+# -------------------------
+def build_keyword_prompt(context: dict) -> str:
+    payload = json.dumps(context, indent=2, default=str)
+
+    return f"""
+You are an expert Google Ads performance strategist for a preschool business.
+
+Campaign: {context.get("campaign_name", "ALL CAMPAIGNS")}
+
+Your job:
+Analyze keyword performance and identify:
+- Winning keywords (high intent + conversions)
+- Wasted spend keywords (cost but no conversions)
+- Keywords to scale
+- Keywords to pause or reduce bids
+- Any CTR / conversion anomalies
+
+IMPORTANT:
+- Do NOT assume CRM / SaaS context
+- Assume all data relates to preschool admissions, daycare, or childcare services
+- Think like a preschool marketing expert
+
+DATA:
+{payload}
+
+Return 5 bullet points.
+Each bullet must start with "- ".
+Be direct, business-focused, no intro text.
+"""
+
+
+# -------------------------
+# Main runner
+# -------------------------
+def run_keyword_inspector(dfs: dict, campaign_name: str | None = None) -> str:
+    print("\n🚀 [keyword_inspector] STARTED", flush=True)
+
+    if not dfs or "keywords" not in dfs:
+        return "⚠️ No keyword data available."
+
+    df = dfs["keywords"].copy()
+    df = build_keyword_features(df)
+
+    # optional campaign filter (safe, not destructive)
+    if campaign_name and "campaign_name" in df.columns:
+        df = df[df["campaign_name"] == campaign_name]
+
+    context = {
+        "campaign_name": campaign_name,
+        "keywords": df.to_dict("records")  # FULL DATA given to LLM
+    }
+
+    print("🧠 [keyword_inspector] context built", flush=True)
+
+    prompt = build_keyword_prompt(context)
+    print("✍️ [keyword_inspector] prompt built", flush=True)
+
+    result = generate_explanation(prompt)
+
+    if is_bad_llm_output(result):
+        print("⚠️ [keyword_inspector] LLM fallback triggered", flush=True)
+        return (
+            "- Unable to generate LLM insights right now.\n"
+            "- Check keyword data quality or retry."
+        )
+
+    print("📤 [keyword_inspector] result received", flush=True)
+    return result

app/ads1/search_term_optimizer.py

@@ -43,20 +43,6 @@ def detect_review_terms(df):
 def detect_scaling_terms(df):
     return df[(df["conversions"] > 0)].sort_values("cpa", ascending=True)
 
-# def classify_search_term(dfs: dict) -> dict:
-#     df = dfs["search_terms"].copy()
-#     df = build_search_term_features(df)
-
-#     negatives = detect_negative_terms(df)
-#     review = detect_review_terms(df)
-#     winners = detect_scaling_terms(df)
-
-#     return {
-#         "negative_keywords": negatives.to_dict("records"),
-#         "review_terms": review.to_dict("records"),
-#         "winning_terms": winners.head(10).to_dict("records"),
-#     }
-
 def build_search_optimizer_prompt(context: dict) -> str:
     # Convert data to clean JSON string strings for better LLM readability
     neg_json = json.dumps(context['negative_keywords'], indent=2)