Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,266 +1,50 @@
-# app.py
-"""
-Personal Finance Chatbot (Streamlit + IBM watsonx + Watson Assistant)
-- Single-file demo app
-- Requirements: streamlit, pandas, python-dotenv, ibm-watsonx-ai, ibm-watson
-- Set env vars before running (see README block below)
-"""
-
-import os
-import json
-import pandas as pd
-import streamlit as st
-from dotenv import load_dotenv
-from datetime import datetime
-from typing import Dict, Any, List
-
-# IBM SDK imports
-try:
-    from ibm_watsonx_ai import APIClient, Credentials
-except Exception:
-    APIClient = None
-try:
-    from ibm_watson import AssistantV2
-    from ibm_cloud_sdk_core.authenticators import IAMAuthenticator
-except Exception:
-    AssistantV2 = None
-
-# Load env vars from .env if present
-load_dotenv()
-
-# -------------------------
-# Configuration / ENV VARS
-# -------------------------
-WATSONX_API_KEY = os.getenv("WATSONX_API_KEY")      # watsonx IAM apikey
-WATSONX_URL = os.getenv("WATSONX_URL")              # watsonx url/endpoint (e.g. https://us-south.ml.cloud.ibm.com)
-WATSONX_MODEL_ID = os.getenv("WATSONX_MODEL_ID", "ibm/granite-13b-instruct-v2")  # default example model
-
-ASSISTANT_APIKEY = os.getenv("ASSISTANT_APIKEY")    # Watson Assistant api key (optional)
-ASSISTANT_URL = os.getenv("ASSISTANT_URL")          # Watson Assistant url (optional)
-ASSISTANT_ID = os.getenv("ASSISTANT_ID")            # Assistant ID (if using dialog)
-
-# Minimal checks
-if APIClient is None:
-    st.warning("ibm-watsonx-ai not installed. Install with: pip install ibm-watsonx-ai")
-if AssistantV2 is None:
-    st.info("ibm-watson (Assistant) client not installed or optional. Install: pip install ibm-watson")
-
-# -------------------------
-# Helper: WatsonX client
-# -------------------------
-def make_watsonx_client() -> Any:
-    """Create and return a watsonx APIClient or None if not configured"""
-    if not (WATSONX_API_KEY and WATSONX_URL):
-        return None
-    credentials = Credentials(url=WATSONX_URL, api_key=WATSONX_API_KEY)
-    client = APIClient(credentials=credentials)
-    return client
-
-def watsonx_generate(client: Any, prompt: str, model_id: str = None, max_tokens: int = 512) -> str:
-    """
-    Generate text using watsonx foundation model via ibm_watsonx_ai client.
-    This function uses ModelInference utilities available on the client.
-    """
-    if client is None:
-        return "Watsonx client is not configured. Please set WATSONX_API_KEY and WATSONX_URL."
-
-    model_id = model_id or WATSONX_MODEL_ID
-    # Use the client's models or model inference helper:
-    # many SDK variations exist; below uses a common pattern (generate_text).
-    try:
-        model_inference = client.model_inference(model_id=model_id)
-        # simple call; more params (temperature, top_k, max_output_tokens) can be passed via params
-        generated = model_inference.generate_text(prompt=prompt, max_output_tokens=max_tokens)
-        # returned type can be dict/json or string depending on sdk version
-        if isinstance(generated, (dict, list)):
-            # attempt to extract textual content
-            text = json.dumps(generated)  # fallback representation
-        else:
-            text = str(generated)
-        return text
-    except Exception as e:
-        return f"Error calling watsonx generate: {e}"
-
-# -------------------------
-# Optional: Watson Assistant helper (for intents / dialog)
-# -------------------------
-def make_assistant_client():
-    if not (ASSISTANT_APIKEY and ASSISTANT_URL):
-        return None
-    auth = IAMAuthenticator(ASSISTANT_APIKEY)
-    assistant = AssistantV2(version="2024-01-01", authenticator=auth)
-    assistant.set_service_url(ASSISTANT_URL)
-    return assistant
-
-# -------------------------
-# Budget parsing & analytics
-# -------------------------
-def parse_transactions_csv(uploaded_file) -> pd.DataFrame:
-    """
-    Expect CSV with columns: date, description, amount, category (category optional).
-    Returns dataframe with normalized date and numeric amount.
-    """
-    df = pd.read_csv(uploaded_file)
-    # basic normalization
-    if "date" in df.columns:
-        df["date"] = pd.to_datetime(df["date"], errors="coerce")
-    if "amount" in df.columns:
-        df["amount"] = pd.to_numeric(df["amount"], errors="coerce")
-    # create category if missing (simple rule-based)
-    if "category" not in df.columns:
-        df["category"] = df["description"].fillna("").str.lower().apply(guess_category_from_desc)
-    df = df.dropna(subset=["amount"])
-    return df
-
-def guess_category_from_desc(desc: str) -> str:
-    desc = (desc or "").lower()
-    if any(k in desc for k in ["uber", "ola", "cab", "taxi"]):
-        return "transport"
-    if any(k in desc for k in ["grocery", "walmart", "bigbasket", "grocer"]):
-        return "groceries"
-    if any(k in desc for k in ["rent", "apartment", "house"]):
-        return "rent"
-    if any(k in desc for k in ["netflix", "spotify", "prime", "hulu"]):
-        return "entertainment"
-    if any(k in desc for k in ["salary", "pay", "deposit"]):
-        return "income"
-    return "other"
-
-def budget_summary(df: pd.DataFrame) -> Dict[str, Any]:
-    """
-    Generate a simple summary: total income, total expenses, top categories.
-    """
-    income = df[df["amount"] > 0]["amount"].sum()
-    expenses = -df[df["amount"] < 0]["amount"].sum()  # amounts might be negative for expenses
-    by_cat = df.groupby("category")["amount"].sum().sort_values()
-    top_exp = by_cat[by_cat < 0].sort_values().head(5) * -1
-    return {
-        "total_income": float(income),
-        "total_expenses": float(expenses),
-        "net_savings": float(income - expenses),
-        "top_expense_categories": top_exp.to_dict()
-    }
-
-# -------------------------
-# Prompt engineering helpers
-# -------------------------
-def build_prompt(user_question: str, demographic: str, budget_summary_text: str = "") -> str:
-    """
-    Build a watsonx prompt which adjusts tone and complexity based on demographic.
-    demographic: "student" or "professional"
-    """
-    tone = "friendly, simple, and educational" if demographic.lower() == "student" else "concise, professional, actionable"
-    complexity = "use short, clear sentences and examples" if demographic.lower() == "student" else "use precise financial language; include bullet recommendations"
-    prompt = (
-        f"You are a helpful personal finance assistant. Adopt a {tone} tone and {complexity}.\n\n"
-        f"Context (if any):\n{budget_summary_text}\n\n"
-        f"User question: {user_question}\n\n"
-        f"Provide:\n1) Short answer to the user's question.\n2) A 3-point actionable plan or suggestion.\n3) If the question is budget-related, give 1 quick saving tip.\n\nAnswer:"
-    )
-    return prompt
-
-# -------------------------
-# Streamlit UI
-# -------------------------
-st.set_page_config(page_title="Personal Finance Chatbot", layout="wide")
-st.title("💬 Personal Finance Chatbot — Savings, Taxes, Investments (IBM watsonx)")
-
-# Sidebar: upload transactions and settings
-st.sidebar.header("Data & Settings")
-uploaded = st.sidebar.file_uploader("Upload transactions CSV (date,description,amount,category optional)", type=["csv"])
-demographic = st.sidebar.selectbox("User type (affects tone & complexity)", ["student", "professional"])
-model_choice = st.sidebar.text_input("watsonx model id", value=WATSONX_MODEL_ID)
-
-st.sidebar.markdown("**API status**")
-watsonx_client = make_watsonx_client()
-assistant_client = make_assistant_client()
-st.sidebar.write("watsonx configured:", bool(watsonx_client))
-st.sidebar.write("Assistant configured:", bool(assistant_client))
-
-# Load transactions if provided
-tx_df = None
-budget_text = ""
-if uploaded:
-    try:
-        tx_df = parse_transactions_csv(uploaded)
-        st.sidebar.success(f"Loaded {len(tx_df)} transactions")
-        summary = budget_summary(tx_df)
-        # create a short textual budget summary for context to the model
-        budget_text = (
-            f"Total income: {summary['total_income']:.2f}. "
-            f"Total expenses: {summary['total_expenses']:.2f}. "
-            f"Net savings: {summary['net_savings']:.2f}. "
-            f"Top expense categories: {', '.join([f'{k}: {v:.2f}' for k,v in summary['top_expense_categories'].items()])}."
-        )
-    except Exception as e:
-        st.sidebar.error(f"Failed to parse CSV: {e}")
-
-# Main: chat area
-if "chat_messages" not in st.session_state:
-    st.session_state.chat_messages = []
-
-chat_col, info_col = st.columns([3,1])
-with chat_col:
-    st.subheader("Chat")
-    # display history
-    for msg in st.session_state.chat_messages:
-        if msg["role"] == "user":
-            st.chat_message("user").write(msg["content"])
+# Add a FINANCE_KEYWORDS list at the top
+FINANCE_KEYWORDS = [
+    "finance", "money", "budget", "spend", "expense", "investment", "invest",
+    "mutual fund", "savings", "loan", "credit", "debit", "stock", "tax",
+    "insurance", "emi", "pay", "salary", "income", "expense", "roi",
+    "interest", "dividend", "bond", "rate", "portfolio", "wealth", "goal",
+    "sip", "fd", "rd", "fixed deposit", "asset", "liability", "capital"
+]
+
+def is_finance_related(text):
+    text_l = text.lower()
+    # Keywords
+    if any(word in text_l for word in FINANCE_KEYWORDS):
+        return True
+    # Numbers (strip commas, dots, %)
+    if any(char.isdigit() for char in text):
+        return True
+    return False
+
+# In your chat input section, filter before sending to the AI
+with col_chat:
+    st.subheader("🗣️ Ask about finance or numbers only")
+    for turn in st.session_state.chat_history:
+        with st.chat_message(turn["role"]):
+            st.markdown(turn["content"])
+    user_msg = st.chat_input("Type your finance/numbers-related question…")
+    if user_msg:
+        if not is_finance_related(user_msg):
+            assistant_message = "Sorry, I can only answer questions related to finance or numbers. Please rephrase your query."
+            st.session_state.chat_history.append({"role": "assistant", "content": assistant_message})
+            with st.chat_message("assistant"):
+                st.markdown(assistant_message)
         else:
-            st.chat_message("assistant").write(msg["content"])
-
-    user_input = st.chat_input("Ask about savings, taxes, budgets, or investments...")
-    if user_input:
-        # show the user's message immediately
-        st.session_state.chat_messages.append({"role":"user", "content": user_input})
-        # Build prompt for watsonx
-        prompt = build_prompt(user_input, demographic, budget_summary_text=budget_text)
-        with st.spinner("Generating response from watsonx..."):
-            if watsonx_client:
-                response_text = watsonx_generate(watsonx_client, prompt=prompt, model_id=model_choice, max_tokens=512)
-            else:
-                # fallback heuristic answer if watsonx not configured
-                response_text = (
-                    "Watsonx not configured. (Set WATSONX_API_KEY and WATSONX_URL.)\n\n"
-                    "Quick tips: 1) Track monthly expenses; 2) Build 3-6 months emergency fund; 3) Automate savings."
-                )
-        # Append assistant response
-        st.session_state.chat_messages.append({"role":"assistant", "content": response_text})
-        st.experimental_rerun()
-
-with info_col:
-    st.subheader("Quick Tools")
-    if tx_df is not None:
-        st.markdown("**Budget snapshot**")
-        st.write(budget_text)
-        if st.button("Show top expenses table"):
-            top = tx_df.groupby("category")["amount"].sum().abs().sort_values(ascending=False).reset_index()
-            top.columns = ["category","total_spent"]
-            st.table(top.head(10))
-    else:
-        st.info("Upload transactions to enable budget summaries & insights")
-
-# Footer: example prompts & environment instructions
-st.markdown("---")
-st.markdown("**Example prompts**")
-st.markdown("- `How much should I save monthly given my income is ₹50,000?`")
-st.markdown("- `Suggest tax-saving instruments suitable for a young professional in India.`")
-st.markdown("- `I spent too much on food. How can I cut dining expenses by 20%?`")
-
-st.markdown("---")
-st.markdown("**Setup / Requirements**")
-st.code("""
-pip install streamlit pandas python-dotenv ibm-watsonx-ai ibm-watson
-# Environment variables (example)
-export WATSONX_API_KEY='your_watsonx_api_key'
-export WATSONX_URL='https://<region>.ml.cloud.ibm.com'
-export WATSONX_MODEL_ID='ibm/granite-13b-instruct-v2'
-# Optional (Assistant)
-export ASSISTANT_APIKEY='your_assistant_apikey'
-export ASSISTANT_URL='https://api.<region>.assistant.watson.cloud.ibm.com'
-export ASSISTANT_ID='your_assistant_id'
-# Run
-streamlit run app.py
-""", language="bash")
-
-st.caption("This is a prototype demo — for production, add secure secret handling, input validation, rate-limiting, logging, and robust error handling.")
+            st.session_state.chat_history.append({"role": "user", "content": user_msg})
+            intent = detect_intent(user_msg)
+            sys_prompt = (
+                "You are a finance and numbers-only AI assistant. "
+                "If asked anything outside of finance or numbers, politely refuse to answer."
+                "\n\n" + build_system_prompt(profile)
+            )
+            usr_prompt = craft_user_prompt(user_msg, intent, summary)
+            final_prompt = sys_prompt + "\n\n" + usr_prompt
+            with st.chat_message("assistant"):
+                with st.spinner(f"Thinking with {provider.name}…"):
+                    try:
+                        ai = provider.generate(final_prompt, max_tokens=768)
+                    except Exception as e:
+                        ai = f"Provider error: {e}\nFalling back to heuristic guidance.\n" + usr_prompt
+                st.markdown(ai)
+            st.session_state.chat_history.append({"role": "assistant", "content": ai})