Update app.py

app.py

@@ -1,3 +1,20 @@
+# main.py
+"""
+Financial Health Score Service (FastAPI)
+
+Features:
+- Fetches user budgets and last-30-day transactions from MongoDB
+- Looks up transaction currency (from `currencies` collection) and embeds currency code
+- Builds a careful prompt for OpenAI (gpt-4o-mini) that instructs usage of currency code
+- Calls OpenAI and reliably extracts JSON output
+- Returns {"userId", "score", "explanation"}
+
+IMPORTANT:
+- Ensure MONGO_URI and OPENAI_API_KEY are set in your environment.
+- The `currencies` collection name is assumed to be "currencies".
+- This file uses the OpenAI Python client (OpenAI(api_key=...)) per your earlier setup.
+"""
+
 import json
 import os
 import re
@@ -31,6 +48,7 @@ if default_db is None:
 
 budget_collection = default_db["budget"]
 transaction_collection = default_db["transactions"]
+currencies_collection = default_db["currencies"]  # <-- currencies collection
 
 # OpenAI client
 openai = OpenAI(api_key=OPENAI_API_KEY)
@@ -49,7 +67,6 @@ class ScoreRequest(BaseModel):
 # ===========================
 # HELPERS
 # ===========================
-
 def safe_number(value):
     try:
         return float(value)
@@ -89,6 +106,9 @@ def normalize_budgets(budgets):
 
 
 def normalize_transactions(transactions):
+    """
+    Trim transactions and attach currency code (e.g., 'EUR', 'INR', 'USD') when possible.
+    """
     trimmed = []
     for txn in transactions:
         date_value = txn.get("date")
@@ -98,18 +118,55 @@ def normalize_transactions(transactions):
         else:
             date_str = None
 
+        # ---- Currency lookup ----
+        currency_code = None
+        currency_id = txn.get("currency")
+
+        try:
+            # currency may be stored as an ObjectId already; handle strings too
+            if isinstance(currency_id, ObjectId):
+                currency_doc = currencies_collection.find_one({"_id": currency_id})
+            elif isinstance(currency_id, dict) and "$oid" in currency_id:
+                # sometimes the raw export contains {"$oid": "..."}
+                try:
+                    currency_doc = currencies_collection.find_one({"_id": ObjectId(currency_id["$oid"])})
+                except Exception:
+                    currency_doc = None
+            elif isinstance(currency_id, str):
+                try:
+                    currency_doc = currencies_collection.find_one({"_id": ObjectId(currency_id)})
+                except Exception:
+                    currency_doc = currencies_collection.find_one({"code": currency_id}) or currencies_collection.find_one({"currency": currency_id})
+            else:
+                currency_doc = None
+
+            if currency_doc:
+                # Option A: use currency code only (e.g., "EUR")
+                currency_code = currency_doc.get("code") or currency_doc.get("currency")
+
+        except Exception:
+            currency_code = None
+
         trimmed.append({
             "type": txn.get("type"),
             "amount": safe_number(txn.get("amount")),
+            "currency": currency_code,  # <-- added
             "date": date_str,
         })
     return trimmed
 
 
 def score_prompt(budgets, transactions):
+    # We instruct the model to use currency code when mentioning amounts (Option A)
     return f"""
-You are a financial wellness expert. Using any available budgets and recent transactions below,
-rate the user's financial health on a scale of 0–100.
+You are a succinct financial wellness expert. Using the budgets and last 30 days of transactions below,
+rate the user's financial health on a scale from 0 to 100 (higher is better).
+
+IMPORTANT:
+- When referring to monetary amounts, ALWAYS prefix with the currency code if available.
+  Example: "EUR 10,000", "INR 5,000", "USD 200".
+- If a transaction has no currency code, you may use the number only (e.g., 1000).
+- Keep the explanation short (one or two sentences) and directly related to budgets and transactions.
 
 Budgets:
 {json.dumps(normalize_budgets(budgets), indent=2)}
@@ -117,63 +174,69 @@ Budgets:
 Transactions (last 30 days):
 {json.dumps(normalize_transactions(transactions), indent=2)}
 
-Respond ONLY with:
-{{"score": number, "explanation": "short explanation"}}
+Respond only with valid JSON, nothing else, using this exact shape:
+{{ "score": number, "explanation": "short explanation" }}
 """
-
+    
 
 def extract_json_payload(text):
     """Extract JSON from plain text or fenced code blocks."""
     trimmed = (text or "").strip()
 
+    # try fenced json block first
     fenced = re.search(r"```(?:json)?\s*([\s\S]*?)```", trimmed)
     if fenced:
         return json.loads(fenced.group(1).strip())
 
+    # try to find first { ... } substring
+    first_obj = re.search(r"(\{[\s\S]*\})", trimmed)
+    if first_obj:
+        return json.loads(first_obj.group(1))
+
+    # last resort: direct JSON load
     return json.loads(trimmed)
 
 
 # ===========================
-# NEW BULLETPROOF OPENAI EXTRACTOR
+# BULLETPROOF OPENAI EXTRACTOR
 # ===========================
-
 def extract_openai_text(response):
     """
-    Bulletproof extractor for OpenAI SDK v1.x ChatCompletion responses.
-    Handles:
-      - message.content (real JSON string)
-      - ChatCompletionMessage(content='...') object repr (your case)
-      - strings
-      - lists
+    Robust extractor for OpenAI SDK responses.
+    Handles several possible message wrappers and returns the assistant text.
     """
     try:
-        msg = response.choices[0].message
+        # Best-effort to access nested choice message content
+        choices = getattr(response, "choices", None) or response.get("choices") if isinstance(response, dict) else None
+        if not choices:
+            # fallback: maybe response is a dict-like structure
+            return str(response)
+
+        msg = choices[0].get("message") if isinstance(choices[0], dict) else getattr(choices[0], "message", None)
+        if not msg:
+            return str(choices[0])
+
+        # If message exposes 'content'
+        if isinstance(msg, dict) and "content" in msg:
+            return msg["content"].strip()
+        if hasattr(msg, "content"):
+            return msg.content.strip()
+
+        # If message is a repr like ChatCompletionMessage(content='...'), extract via regex
+        msg_str = str(msg)
+        match = re.search(r"content='([\s\S]*?)'", msg_str)
+        if match:
+            return match.group(1).strip()
+
+        # fallback
+        return msg_str.strip()
     except Exception:
-        raise HTTPException(status_code=502, detail="OpenAI returned no valid choices")
-
-    # Case 1: content exists normally
-    if hasattr(msg, "content"):
-        return msg.content.strip()
-
-    # Case 2: msg is dict containing 'content'
-    if isinstance(msg, dict) and "content" in msg:
-        return str(msg["content"]).strip()
-
-    # Case 3: msg is a Python repr:
-    # ChatCompletionMessage(content='{"score":...}', role='assistant')
-    msg_str = str(msg)
-    match = re.search(r"content='([\s\S]*?)'", msg_str)
-    if match:
-        return match.group(1).strip()
-
-    # Fallback
-    return msg_str.strip()
+        return str(response)
 
 
 # ===========================
 # ROUTES
 # ===========================
-
 @app.get("/")
 def root():
     return {"status": "Financial Health Score service is running"}
@@ -187,7 +250,7 @@ def financial_score(payload: ScoreRequest):
     except Exception:
         raise HTTPException(status_code=400, detail="Invalid userId")
 
-    # Fetch budgets
+    # Fetch budgets (all budgets created by this user)
     budgets = list(budget_collection.find({"createdBy": user_id}))
 
     # Fetch last 30 days transactions
@@ -199,11 +262,12 @@ def financial_score(payload: ScoreRequest):
         }).sort("date", -1).limit(100)
     )
 
+    # If neither budgets nor recent transactions exist -> score 0
     if not budgets and not transactions:
         return {
             "userId": payload.userId,
             "score": 0,
-            "explanation": "No budgets or transactions found."
+            "explanation": "No budgets or recent transactions found."
         }
 
     prompt = score_prompt(budgets, transactions)
@@ -218,17 +282,16 @@ def financial_score(payload: ScoreRequest):
     except Exception as exc:
         raise HTTPException(status_code=502, detail=f"OpenAI request failed: {exc}")
 
-    # Extract text safely
     model_output = extract_openai_text(response)
 
-    # Parse JSON
+    # Parse JSON payload from model output
     try:
         parsed = extract_json_payload(model_output)
     except Exception:
         raise HTTPException(
             status_code=502,
             detail={
-                "error": "Unable to parse OpenAI response",
+                "error": "Unable to parse OpenAI response as JSON",
                 "rawResponse": model_output
             },
         )
@@ -243,7 +306,7 @@ def financial_score(payload: ScoreRequest):
             }
         )
 
-    # Score limits
+    # Clamp score to 0..100
     try:
         score_val = int(float(parsed["score"]))
         score_val = max(0, min(100, score_val))
@@ -255,3 +318,4 @@ def financial_score(payload: ScoreRequest):
         "score": score_val,
         "explanation": parsed["explanation"]
     }
+