Update src/streamlit_app.py

src/streamlit_app.py

@@ -2,18 +2,15 @@ import streamlit as st
 import pandas as pd
 import matplotlib.pyplot as plt
 import google.generativeai as genai
-import json, re
-import os
-from datetime import datetime
+import json, os, re
 
 # ====== CONFIG ======
 st.set_page_config(page_title="ECL Risk Analyzer", layout="wide")
 genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
 
-# ====== FUNCTIONS ======
+# ====== HELPERS ======
 @st.cache_data
 def process_loan_data(df: pd.DataFrame):
-    """Compute PD, LGD, EAD, and ECL by loan_intent."""
     df = df.dropna(subset=["loan_intent", "credit_score", "loan_amnt", "loan_status"])
     df["loan_status"] = df["loan_status"].astype(int)
     group = df.groupby("loan_intent")
@@ -23,21 +20,17 @@ def process_loan_data(df: pd.DataFrame):
     ecl_seg = pd_seg * lgd_seg * ead_seg
     ecl_df = pd.concat([pd_seg, lgd_seg, ead_seg, ecl_seg], axis=1)
     ecl_df.columns = ["PD", "LGD", "EAD", "ECL"]
-    ecl_df = ecl_df.reset_index()
-    return ecl_df
-
-import re, json
+    return ecl_df.reset_index()
 
 def get_gemini_decision(segment, pd_val, lgd_val, ead_val, ecl_val):
-    """Gemini-backed risk decision, with hardened JSON cleanup."""
+    """Gemini-backed risk decision, single-segment call with robust parsing."""
     model = genai.GenerativeModel("gemini-2.0-flash-lite")
 
     system_prompt = (
         "You are a financial risk advisor. "
-        "Return only JSON, never markdown. "
-        'Format: {"action":"increase_interest"|"reduce_disbursement"|"maintain","rationale":"string","confidence":float}'
+        "Return only JSON. "
+        'Schema: {"action":"increase_interest"|"reduce_disbursement"|"maintain","rationale":"string","confidence":float}'
     )
-
     user_prompt = f"""
 Segment: {segment}
 PD: {pd_val:.3f}
@@ -59,96 +52,61 @@ Respond with one JSON object only.
             generation_config={"temperature": 0.1}
         )
         text = resp.text.strip()
-
-        # --- Strip Markdown wrappers like ```json ... ```
         text = re.sub(r"^```json", "", text)
         text = re.sub(r"^```", "", text)
         text = re.sub(r"```$", "", text)
-        text = text.strip()
-
-        # --- Extract only JSON substring ---
         match = re.search(r"\{.*\}", text, re.DOTALL)
         if match:
             text = match.group(0)
-
-        # --- Load and validate ---
         data = json.loads(text)
-        if not isinstance(data, dict):
-            raise ValueError("Parsed non-dict JSON")
-
-        for k in ["action", "rationale", "confidence"]:
-            data.setdefault(k, None)
         return data
-
     except Exception as e:
-        # Log what Gemini returned for debugging
         st.warning(f"⚠️ Gemini output parse failed: {e}")
         st.text_area("Raw Gemini output", value=resp.text if 'resp' in locals() else "No response", height=150)
         return {"action": "maintain", "rationale": "Fallback - parse failure", "confidence": 0.0}
 
 # ====== UI ======
 st.title("📊 Expected Credit Loss (ECL) Risk Dashboard")
-st.write("Upload your **bank loan dataset** to compute segment-level Expected Credit Loss (ECL) and get AI-driven recommendations.")
+st.write("Upload your **loan dataset**, review segment-level ECL metrics, and analyze one segment at a time with Gemini.")
 
 uploaded = st.file_uploader("Upload CSV dataset", type=["csv"])
 
 if uploaded:
     df = pd.read_csv(uploaded)
-    st.success("Dataset loaded successfully.")
-    st.dataframe(df.head())
-
     ecl_df = process_loan_data(df)
-    st.subheader("Segment-level ECL Summary")
+    st.success("Dataset processed successfully.")
     st.dataframe(ecl_df, use_container_width=True, hide_index=True)
 
-    # --- Visualization: ECL by segment ---
-    st.subheader("ECL by Segment")
-    fig, ax = plt.subplots(figsize=(8, 4))
-    ax.bar(ecl_df["loan_intent"], ecl_df["ECL"])
-    ax.set_xlabel("Segment")
-    ax.set_ylabel("ECL")
-    ax.set_title("Expected Credit Loss per Segment")
-    plt.xticks(rotation=45)
-    st.pyplot(fig)
-
-    # --- Visualization: PD by segment ---
-    st.subheader("PD by Segment")
-    fig2, ax2 = plt.subplots(figsize=(8, 4))
-    ax2.bar(ecl_df["loan_intent"], ecl_df["PD"], color="gray")
-    ax2.set_xlabel("Segment")
-    ax2.set_ylabel("PD")
-    ax2.set_title("Probability of Default (PD) per Segment")
-    plt.xticks(rotation=45)
-    st.pyplot(fig2)
-
-    # --- AI Decision Section ---
-    st.subheader("AI Recommendations (Gemini)")
-    decisions = []
-    for _, row in ecl_df.iterrows():
-        decision = get_gemini_decision(row["loan_intent"], row["PD"], row["LGD"], row["EAD"], row["ECL"])
-        decisions.append({
-            "Segment": row["loan_intent"],
-            "Action": decision["action"],
-            "Rationale": decision["rationale"],
-            "Confidence": decision["confidence"],
-            "ECL": row["ECL"],
-            "PD": row["PD"]
-        })
-    result_df = pd.DataFrame(decisions)
-    result_df["Timestamp"] = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-    st.dataframe(result_df, use_container_width=True, hide_index=True)
-
-    # --- Plot action summary ---
-    st.subheader("Recommended Actions Distribution")
-    fig3, ax3 = plt.subplots(figsize=(6, 4))
-    action_counts = result_df["Action"].value_counts()
-    ax3.pie(action_counts, labels=action_counts.index, autopct="%1.1f%%", startangle=140)
-    ax3.set_title("Recommended Actions per Segment")
-    st.pyplot(fig3)
-
-    # Option to export report
-    csv_out = result_df.to_csv(index=False).encode("utf-8")
-    st.download_button("Download ECL + Decision Report", csv_out, "ECL_Decisions.csv", "text/csv")
+    # --- Visual overview ---
+    col1, col2 = st.columns(2)
+    with col1:
+        st.subheader("ECL by Segment")
+        fig, ax = plt.subplots(figsize=(6, 3))
+        ax.bar(ecl_df["loan_intent"], ecl_df["ECL"])
+        ax.set_xlabel("Segment"); ax.set_ylabel("ECL")
+        plt.xticks(rotation=45)
+        st.pyplot(fig)
+    with col2:
+        st.subheader("PD by Segment")
+        fig2, ax2 = plt.subplots(figsize=(6, 3))
+        ax2.bar(ecl_df["loan_intent"], ecl_df["PD"], color="gray")
+        ax2.set_xlabel("Segment"); ax2.set_ylabel("PD")
+        plt.xticks(rotation=45)
+        st.pyplot(fig2)
+
+    # --- Segment selection ---
+    st.subheader("Analyze Specific Segment")
+    segments = ecl_df["loan_intent"].unique().tolist()
+    selected_segment = st.selectbox("Choose a segment:", segments)
+
+    row = ecl_df[ecl_df["loan_intent"] == selected_segment].iloc[0]
+    st.write(f"**PD:** {row.PD:.3f} | **LGD:** {row.LGD:.3f} | **EAD:** {row.EAD:,.0f} | **ECL:** {row.ECL:,.0f}")
+
+    if st.button("Generate Gemini Decision"):
+        with st.spinner("Querying Gemini..."):
+            decision = get_gemini_decision(row["loan_intent"], row["PD"], row["LGD"], row["EAD"], row["ECL"])
+        st.success("Gemini Decision:")
+        st.json(decision)
 
 else:
-    st.info("Upload your CSV file to begin analysis.")
+    st.info("Upload a CSV file to begin.")