huggingfaceModel.py

import os
from huggingface_hub import InferenceClient

def generate_dropout_insights(input_values, risk_factors, label):
    """
    Generates structured dropout risk insights using AI, ensuring Unicode-safe text for PDFs.
    """
    excluded_fields = {"roll_no", "semester", "degree"}  # Fields to exclude from analysis
    
    # Filter only relevant input values
    filtered_input_values = {key: value for key, value in input_values.items() if key not in excluded_fields}
    
    # Map input values to their corresponding risk factors
    risk_mapping = {key: risk for key, risk in zip(filtered_input_values.keys(), risk_factors)}
    
    # Construct message for AI model
    formatted_input = "\n".join([f"{key}: {value} (Risk: {risk})" for key, (value, risk) in zip(filtered_input_values.keys(), zip(filtered_input_values.values(), risk_factors))])
    
    messages = [
        {
            "role": "user",
            "content": f"""
            Generate a detailed dropout risk analysis report based on the following data:
            
            {formatted_input}
            
            Format the response as:
            
            ## Analysis of Student Dropout Risk
            
            Overall Risk Level: {label}
            
            ### 🟢 Strengths:
            List areas where the student is performing well:
            {{strengths}}
            
            ### 🔴 Concerns:
            Identify factors that increase dropout risk:
            {{concerns}}
            
            ### 📌 Action Plan for Improvement:
            Provide targeted **solutions** based on the student's **weak areas**:
            {{recommendations}}
            
            ## 📝 Final Recommendation:
            {{overall_recommendation}}
            """
        }
    ]
    
    client = InferenceClient(
        provider="nebius",
        api_key=os.getenv("HUGGINGFACE_API_TOKEN")
    )
    
    stream = client.chat.completions.create(
        model="deepseek-ai/DeepSeek-R1", 
        messages=messages, 
        temperature=0.5,
        max_tokens=2048,
        top_p=0.7,
        stream=True
    )
    
    # Collect and return the generated report
    report = ""
    for chunk in stream:
        report += chunk.choices[0].delta.content
    
    return report