Update analytics/ai_assistant.py

analytics/ai_assistant.py

@@ -123,6 +123,77 @@ def build_portfolio_context(df: pd.DataFrame, as_of_month: str | None = None, se
 
     return "\n".join(lines)
 
+def build_calendar_performance_context(df: pd.DataFrame, as_of_month: str | None = None):
+    if not as_of_month or as_of_month == "All":
+        return ""
+
+    month_df = _filter_by_month(df, as_of_month)
+    if month_df.empty:
+        return f"No data found for {as_of_month}."
+
+    lines = [f"Calendar month snapshot for {as_of_month}:"]
+    if "balance" in month_df.columns:
+        lines.append(f"  - Total balance: {month_df['balance'].sum(skipna=True):.0f}")
+    lines.append(f"  - Total accounts: {month_df['account_id'].nunique() if 'account_id' in month_df.columns else 0}")
+    if "dpd" in month_df.columns:
+        lines.append(f"  - Accounts with dpd>=30: {month_df.loc[month_df['dpd'].fillna(0) >= 30, 'account_id'].nunique() if 'account_id' in month_df.columns else 0}")
+
+    for metric in RISK_METRICS:
+        try:
+            view = generate_metric_view(month_df, metric_name=metric, group_col=None)
+            rate_col = [c for c in view.columns if "rate" in c.lower()][0]
+            metric_rate = view.loc[0, rate_col]
+            lines.append(f"  - {metric} rate: {_fmt_pct(metric_rate)}")
+        except Exception:
+            continue
+
+    return "\n".join(lines)
+
+
+def build_vintage_performance_context(df: pd.DataFrame):
+    if "booking_vintage" not in df.columns:
+        return ""
+
+    lines = ["Vintage performance summary:"]
+    for metric in RISK_METRICS:
+        try:
+            view = generate_metric_view(df, metric_name=metric, group_col="booking_vintage")
+            rate_col = [c for c in view.columns if "rate" in c.lower()][0]
+            if view.empty:
+                continue
+            top = view.sort_values(rate_col, ascending=False).head(2)
+            bottom = view.sort_values(rate_col, ascending=True).head(1)
+            lines.append(
+                f"  - {metric}: highest risk vintage {top.iloc[0]['booking_vintage']} at {_fmt_pct(top.iloc[0][rate_col])}, "
+                f"lowest risk vintage {bottom.iloc[0]['booking_vintage']} at {_fmt_pct(bottom.iloc[0][rate_col])}."
+            )
+        except Exception:
+            continue
+
+    return "\n".join(lines)
+
+
+def build_context_for_question(df: pd.DataFrame, as_of_month: str | None = None, segment: str | None = None, question: str | None = None):
+    sections = [build_portfolio_context(df, as_of_month=as_of_month, segment=segment)]
+    q = (question or "").lower()
+
+    if "month" in q or "calendar" in q or "as of" in q:
+        month_context = build_calendar_performance_context(df, as_of_month=as_of_month)
+        if month_context:
+            sections.append(month_context)
+
+    if "vintage" in q or "trend" in q or "booking vintage" in q or "compare" in q:
+        vintage_context = build_vintage_performance_context(df)
+        if vintage_context:
+            sections.append(vintage_context)
+
+    if segment and segment in df.columns:
+        segment_context = build_portfolio_context(df, as_of_month=as_of_month, segment=segment)
+        if segment_context:
+            sections.append(segment_context)
+
+    return "\n\n".join(section for section in sections if section)
+
 
 def _get_inference_client():
     if not HF_TOKEN:
@@ -136,7 +207,7 @@ def _get_inference_client():
 
 
 def generate_ai_answer(question: str, df: pd.DataFrame, as_of_month: str | None = None, segment: str | None = None):
-    summary = build_portfolio_context(df, as_of_month=as_of_month, segment=segment)
+    summary =  build_context_for_question(df, as_of_month=as_of_month, segment=segment, question=question)
     prompt = (
         "You are a senior risk manager assistant responding to portfolio analytics questions. "
         "Use the risk context below and answer the user clearly, describing what is happening, why it is happening, and what actions should be considered. "