submission-FINCOP

README.md

@@ -9,18 +9,53 @@
 
 ## README.md must consist of the following information:
 
-#### Team Name -
-#### Problem Statement - 
-#### Team Leader Email -
+#### Team Name - FINCOP
+#### Problem Statement - Real-Time High Scale Financial Fraud Risk Management
+#### Team Leader Email - [email protected]
 
 ## A Brief of the Prototype:
-  This section must include UML Diagrams and prototype description
+  ![use case](use_case_sbi.png)
 
 ## Tech Stack: 
-   List Down all technologies used to Build the prototype
+   Following technologies were used to Build the prototype:
+     1. Langchain\
+     2. OpenAI API\
+     3. Gradio\
+     4. Pytesseract
 
 ## Step-by-Step Code Execution Instructions:
-  This Section must contain a set of instructions required to clone and run the prototype so that it can be tested and deeply analyzed
+
+  Step 1: Clone the repository using the following command:\
+        ```
+        git clone https://github.com/hindesh-akash/Pitch-to-SBI-Hackathon.git
+        ```
+
+  Step 2: Install the requirements:\
+        ```
+        pip install -r requirements.txt
+        ```
+
+  Step 3: **IMPORTANT**\
+        -> Setup your OpenAI key in the **openai.api_key** variable in "loan_fraud.py" and "upi_fraud_check.py"
+
+  Step 4: Open a new terminal
+
+        To check for UPI fraud, run the following code:
+
+        ```
+        python upi_fraud_check.py
+        ```
+Then in the terminal click on the link generated for local host. Example- Running on local URL:  http://127.0.0.1:7861
+
+To check for Loan Portfolio fraud, run the following code:
+
+        ```
+        python loan_fraud.py
+        ```
+
+Then in the terminal click on the link generated for local host. Example- Running on local URL:  http://127.0.0.1:7861
+
+  **Explore the app!**
 
 ## What I Learned:
-   Write about the biggest learning you had while developing the prototype
+   While developing the prototype I learned that if we do not pay attention to our transactions regarding whom we are paying and by what means, we can easily fall prey to financial fraud. I learned how LLMs can be used to leverage fraud detection and how an OCR model can reduce the human task of detecting whether a loan applicant is a defaulter. Similarily I learned the usage of robust machine learning models to detect credit defaults.

examples.txt

@@ -0,0 +1,11 @@
+loan application:
+
+The loan application seeks funding for the construction of a residential property on a 5000 square feet plot of land . The funds will cover materials, labor, permits, and other essential expenses, resulting\
+ in a valuable and comfortable residence in a prime location.
+
+Details:
+Applicant name: John Doe
+House construction site: New Colony, 100001
+Total cost: 5,000,000
+
+

loan_fraud.py

@@ -0,0 +1,68 @@
+import gradio as gr
+import pytesseract
+import openai
+import datetime
+
+# Set your OpenAI API key
+openai.api_key = "----ENTER YOUR OPENAI API KEY HERE----"
+
+# Define the model
+def get_completion(prompt, model="gpt-3.5-turbo"):
+    messages = [{"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": prompt}]
+    response = openai.ChatCompletion.create(
+        model=model,
+        messages=messages,
+        temperature=0,
+    )
+    return response.choices[0].message["content"]
+
+def analyze_sms_fraud(input_image,loan_application):
+
+    salary_details = pytesseract.image_to_string(input_image)
+
+    prompt_financial_summary = f"""
+        I have extracted the information in triple backticks from an Income Tax Return (ITR) form image.\
+        Suppose you are an expert in Income Tax Return (ITR) audit.\
+        Perform the following tasks:\
+
+            1. Calculate the Gross Salaryfrom given data where the gross salary is the sum of all types of incomes\
+                whether it is taxable or not or it is Retirement benefit or not.\
+
+            2. Make a small financial summary of the applicant based on whatever you have calculated.\
+
+        The data is as follows: ```{salary_details}```
+        """
+
+    financial_summary_applicant = get_completion(prompt_financial_summary)
+
+    fraud_detection_prompt = f"""
+        Suppose you are a loan processor at a bank.\
+        Based on the financial summary given in triple backticks and\
+        loan application details in triple askerisks, detect whether the context of financial summary matches the fact that\
+        the applicant will be able to repay the loan or not. Also note there is no pinode like 100001.\
+        If the applicant is unable to repay the loan, then possibly the applicant is fraud.\
+        If the applicant is a fraud, then output the reason for the same.\
+        And if the applicant is not a fraud, then output the reason for the same that how he/she can possibly repay the loan.\
+
+        Financial summary of applicant: ```{financial_summary_applicant}```
+
+        Loan application details: ***{loan_application}***
+        """
+
+    fraud_detection_result = get_completion(fraud_detection_prompt)
+    return fraud_detection_result
+
+# Define the interface
+input_image = gr.inputs.Image(label="Input ITR Image")
+loan_application = gr.inputs.Textbox(label="Loan Application Details")
+
+iface = gr.Interface(
+    fn=analyze_sms_fraud,
+    inputs=[input_image, loan_application],
+    outputs=gr.outputs.Textbox(label="Fraud Detection Result"),
+    title="Loan Fraud Detection",
+    description="Analyze an ITR form image and loan application details for potential fraud using LangChain and OpenAI's GPT-3."
+)
+
+iface.launch()

upi_fraud_check.py

@@ -0,0 +1,57 @@
+import gradio as gr
+import pytesseract
+import openai
+import datetime
+
+# Set your OpenAI API key
+openai.api_key = "----ENTER YOUR OPENAI API KEY HERE----"
+
+# Define the model
+def get_completion(prompt, model="gpt-3.5-turbo"):
+    messages = [{"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": prompt}]
+    response = openai.ChatCompletion.create(
+        model=model,
+        messages=messages,
+        temperature=0,
+    )
+    return response.choices[0].message["content"]
+
+def analyze_sms_fraud(input_text_or_image):
+
+    sms_text = pytesseract.image_to_string(input_text_or_image)
+
+    # Generate prompt for LangChain
+    fraud_sms_prompt = f"""
+    A user received the SMS message in triple backticks from someone claiming to be their bank,
+    Verify what information is being asked from the user. You can do it in following steps:
+    Step 1: Read the SMS message and identify if there is any the bank name.
+    Step 2: Identify is there is request for any personal information/ request for KYC etc.
+    Step 3: Identify if there is any request for OTP or PIN.
+    Step 4: Identify if there is any link in the SMS message and the link is secure or not.
+    Step 5: Identify if there is any request for money transfer.
+    Based on the above steps, identify if the SMS message is a fraud or not.
+    SMS message: ```{sms_text}```
+    """
+
+    # Get fraud detection result from LangChain
+    detected_fraud = get_completion(fraud_sms_prompt)
+
+    return sms_text, detected_fraud
+
+
+input_image = gr.inputs.Image(label="Input SMS Image")
+
+
+iface = gr.Interface(
+    fn=analyze_sms_fraud,
+    inputs= input_image,
+    outputs=[
+        gr.outputs.Textbox(label="SMS Text"),
+        gr.outputs.Textbox(label="Detected Fraud Result")
+    ],
+    title="SMS Fraud Detection",
+    description="Analyze an SMS message for potential fraud using LangChain and OpenAI's GPT-3."
+)
+
+iface.launch()