app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from simple_salesforce import Salesforce
import os
from dotenv import load_dotenv

# Load environment variables
load_dotenv()

# Required env vars check
required_env_vars = ['SF_USERNAME', 'SF_PASSWORD', 'SF_SECURITY_TOKEN']
missing_vars = [var for var in required_env_vars if not os.getenv(var)]
if missing_vars:
    raise EnvironmentError(f"Missing required environment variables: {missing_vars}")

# Defaults
KPI_FLAG_DEFAULT = os.getenv('KPI_FLAG', 'True') == 'True'
ENGAGEMENT_SCORE_DEFAULT = float(os.getenv('ENGAGEMENT_SCORE', '85.0'))

# Load model and tokenizer (Updated to use distilgpt2)
model_name = "distilgpt2"  # Using distilgpt2 for faster response
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(model_name, low_cpu_mem_usage=True)

if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token
    tokenizer.pad_token_id = tokenizer.convert_tokens_to_ids(tokenizer.pad_token)
model.config.pad_token_id = tokenizer.pad_token_id

# Refined Prompt to generate day-by-day tasks based on milestones
PROMPT_TEMPLATE = """You are an AI assistant for construction supervisors. Given the role, project, milestones, and a reflection log, generate:

1. A Daily Checklist with clear and concise tasks based on the role and milestones. 
   Split the checklist into day-by-day tasks for a specified time period (e.g., one week).
2. Focus Suggestions based on concerns or keywords in the reflection log. Provide at least 2 suggestions.

Inputs:
Role: {role}
Project ID: {project_id}
Milestones: {milestones}
Reflection Log: {reflection}

Output Format:
Checklist (Day-by-Day):
- Day 1: 
   - Task 1
   - Task 2
- Day 2: 
   - Task 1
   - Task 2
...
Suggestions:
- 
"""

# Salesforce Functions
def get_roles_from_salesforce():
    try:
        sf = Salesforce(
            username=os.getenv('SF_USERNAME'),
            password=os.getenv('SF_PASSWORD'),
            security_token=os.getenv('SF_SECURITY_TOKEN'),
            domain=os.getenv('SF_DOMAIN', 'login')
        )
        result = sf.query("SELECT Role__c FROM Supervisor__c WHERE Role__c != NULL")
        return list(set(record['Role__c'] for record in result['records']))
    except Exception as e:
        print(f"⚠️ Error fetching roles: {e}")
        return ["Site Manager", "Safety Officer", "Project Lead"]

def get_supervisor_name_by_role(role):
    try:
        sf = Salesforce(
            username=os.getenv('SF_USERNAME'),
            password=os.getenv('SF_PASSWORD'),
            security_token=os.getenv('SF_SECURITY_TOKEN'),
            domain=os.getenv('SF_DOMAIN', 'login')
        )
        role = role.replace("'", "\\'")
        result = sf.query(f"SELECT Name FROM Supervisor__c WHERE Role__c = '{role}'")
        return [record['Name'] for record in result['records']]
    except Exception as e:
        print(f"⚠️ Error fetching supervisor names: {e}")
        return []

def get_projects_for_supervisor(supervisor_name):
    try:
        sf = Salesforce(
            username=os.getenv('SF_USERNAME'),
            password=os.getenv('SF_PASSWORD'),
            security_token=os.getenv('SF_SECURITY_TOKEN'),
            domain=os.getenv('SF_DOMAIN', 'login')
        )
        supervisor_name = supervisor_name.replace("'", "\\'")
        supervisor_result = sf.query(f"SELECT Id FROM Supervisor__c WHERE Name = '{supervisor_name}' LIMIT 1")
        if supervisor_result['totalSize'] == 0:
            return ""
        supervisor_id = supervisor_result['records'][0]['Id']
        project_result = sf.query(f"SELECT Name FROM Project__c WHERE Supervisor_ID__c = '{supervisor_id}' LIMIT 1")
        return project_result['records'][0]['Name'] if project_result['totalSize'] > 0 else ""
    except Exception as e:
        print(f"⚠️ Error fetching project: {e}")
        return ""

def field_exists(sf, object_name, field_name):
    try:
        obj_desc = getattr(sf, object_name).describe()
        return field_name in [field['name'] for field in obj_desc['fields']]
    except Exception as e:
        print(f"⚠️ Error checking field {field_name}: {e}")
        return False

# New function to generate Salesforce dashboard URL (Visualforce Page)
def generate_salesforce_dashboard_url(supervisor_name, project_id):
    # Use the provided Salesforce Visualforce URL with supervisorName and projectId as parameters
    return f"https://aicoachforsitesupervisors-dev-ed--c.develop.vf.force.com/apex/DashboardPage?supervisorName={supervisor_name}&projectId={project_id}"

# Dashboard button function
def open_dashboard(role, supervisor_name, project_id):
    # Generate dynamic URL based on supervisor and project
    dashboard_url = generate_salesforce_dashboard_url(supervisor_name, project_id)
    return f'<a href="{dashboard_url}" target="_blank" rel="noopener noreferrer" style="font-size:16px;">Open Salesforce Dashboard</a>'

# Generate function
def generate_outputs(role, supervisor_name, project_id, milestones, reflection):
    if not all([role, supervisor_name, project_id, milestones, reflection]):
        return "❗ Please fill all fields.", ""

    prompt = PROMPT_TEMPLATE.format(
        role=role,
        project_id=project_id,
        milestones=milestones,
        reflection=reflection
    )

    inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=512)
    try:
        with torch.no_grad():
            outputs = model.generate(
                inputs['input_ids'],
                max_new_tokens=150,  # Increased max tokens to capture more content
                no_repeat_ngram_size=2,
                do_sample=True,
                top_p=0.9,
                temperature=0.7,
                pad_token_id=tokenizer.pad_token_id
            )
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
    except Exception as e:
        print(f"⚠️ Generation error: {e}")
        return "", ""

    def extract_between(text, start, end):
        s = text.find(start)
        e = text.find(end, s) if end else len(text)
        return text[s + len(start):e].strip() if s != -1 else ""

    # Extract the checklist and suggestions
    checklist = extract_between(result, "Checklist:\n", "Suggestions:")
    suggestions = extract_between(result, "Suggestions:\n", None)

    # If checklist or suggestions are empty, generate fallback content
    if not checklist.strip():
        checklist = generate_fallback_checklist(role, milestones)
    if not suggestions.strip():
        suggestions = generate_fallback_suggestions(reflection)

    return checklist, suggestions

# Fallback generation for checklist and suggestions
def generate_fallback_checklist(role, milestones):
    checklist_items = []

    # If milestones are provided, add them to the checklist directly
    if milestones and milestones.strip():
        kpis = [kpi.strip() for kpi in milestones.split(",")]
        for kpi in kpis:
            checklist_items.append(f"- Ensure progress on {kpi}")
    else:
        checklist_items.append("- Perform daily safety inspection")
    
    return "\n".join(checklist_items)

def generate_fallback_suggestions(reflection):
    suggestions_items = []
    reflection_lower = reflection.lower()
    if "student" in reflection_lower or "learning" in reflection_lower:
        suggestions_items.append("- Ensure students are logging incidents consistently")
        suggestions_items.append("- Provide guidance on timely incident recording")
    if "incident" in reflection_lower:
        suggestions_items.append("- Follow up on reported incidents with corrective actions")
    
    if not suggestions_items:
        suggestions_items.append("- Monitor team coordination")
        suggestions_items.append("- Review safety protocols with the team")
    
    return "\n".join(suggestions_items)

# Interface
def create_interface():
    roles = get_roles_from_salesforce()
    with gr.Blocks(theme="soft") as demo:
        gr.Markdown("## 🧠 AI-Powered Supervisor Assistant")

        with gr.Row():
            role = gr.Dropdown(choices=roles, label="Role")
            supervisor_name = gr.Dropdown(choices=[], label="Supervisor Name")
            project_id = gr.Textbox(label="Project ID", interactive=False)

        milestones = gr.Textbox(label="Milestones (comma-separated KPIs)", placeholder="E.g. Safety training, daily inspection")
        reflection = gr.Textbox(label="Reflection Log", lines=4, placeholder="Any concerns, delays, updates...")

        with gr.Row():
            generate = gr.Button("Generate")
            clear = gr.Button("Clear")
            refresh = gr.Button("🔄 Refresh Roles")
            dashboard_btn = gr.Button("Dashboard")

        checklist_output = gr.Textbox(label="✅ Daily Checklist")
        suggestions_output = gr.Textbox(label="💡 Focus Suggestions")
        dashboard_link = gr.HTML("")

        role.change(fn=lambda r: gr.update(choices=get_supervisor_name_by_role(r)), inputs=role, outputs=supervisor_name)
        supervisor_name.change(fn=get_projects_for_supervisor, inputs=supervisor_name, outputs=project_id)

        generate.click(fn=generate_outputs,
                       inputs=[role, supervisor_name, project_id, milestones, reflection],
                       outputs=[checklist_output, suggestions_output])

        clear.click(fn=lambda: ("", "", "", "", ""), inputs=None,
                    outputs=[role, supervisor_name, project_id, milestones, reflection])

        refresh.click(fn=lambda: gr.update(choices=get_roles_from_salesforce()), outputs=role)

        dashboard_btn.click(fn=open_dashboard, inputs=[role, supervisor_name, project_id], outputs=dashboard_link)

    return demo

if __name__ == "__main__":
    app = create_interface()
    app.launch()