app.py

import streamlit as st
import os
import time
import PyPDF2
from docx import Document
import pandas as pd
from dotenv import load_dotenv
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

# Load environment variables
load_dotenv()

# Avatars and bios
USER_AVATAR = "https://raw.githubusercontent.com/achilela/vila_fofoka_analysis/9904d9a0d445ab0488cf7395cb863cce7621d897/USER_AVATAR.png"
BOT_AVATAR = "https://raw.githubusercontent.com/achilela/vila_fofoka_analysis/991f4c6e4e1dc7a8e24876ca5aae5228bcdb4dba/Ataliba_Avatar.jpg"

ATALIBA_BIO = """
**I am Ataliba Miguel's Digital Twin** 🤖

**Background:**
- 🎓 Mechanical Engineering (BSc)
- ⛽ Oil & Gas Engineering (MSc Specialization)
- 🔧 17+ years in Oil & Gas Industry
- 🔍 Current: Topside Inspection Methods Engineer @ TotalEnergies
- 🤖 AI Practitioner Specialist
- 🚀 Founder of ValonyLabs (AI solutions for industrial corrosion, retail analytics, and KPI monitoring)

**Capabilities:**
- Technical document analysis
- Engineering insights
- AI-powered problem solving
- Cross-domain knowledge integration

Ask me about engineering challenges, AI applications, or industry best practices!
"""

# UI Setup
st.markdown("""
    <style>
    @import url('https://fonts.cdnfonts.com/css/tw-cen-mt');
    * { font-family: 'Tw Cen MT', sans-serif; }
    .st-emotion-cache-1y4p8pa { padding: 2rem 1rem; }
    </style>
""", unsafe_allow_html=True)

st.title("🚀 Ataliba o Agent Nerdx 🚀")

# Sidebar
with st.sidebar:
    st.header("⚡️ Hugging Face Model Loaded")
    st.markdown("Model: amiguel/unsloth_finetune_test")
    uploaded_file = st.file_uploader("Upload technical documents", type=["pdf", "docx", "xlsx", "xlsm"])

# Session state
if "file_context" not in st.session_state:
    st.session_state.file_context = None
if "chat_history" not in st.session_state:
    st.session_state.chat_history = []

# File parser
def parse_file(file):
    try:
        if file.type == "application/pdf":
            reader = PyPDF2.PdfReader(file)
            return "\n".join([page.extract_text() for page in reader.pages])
        elif file.type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document":
            doc = Document(file)
            return "\n".join([para.text for para in doc.paragraphs])
        elif file.type in ["application/vnd.openxmlformats-officedocument.spreadsheetml.sheet", "application/vnd.ms-excel"]:
            df = pd.read_excel(file)
            return df.to_string()
    except Exception as e:
        st.error(f"Error processing file: {str(e)}")
        return None

# Process file
if uploaded_file and not st.session_state.file_context:
    st.session_state.file_context = parse_file(uploaded_file)
    if st.session_state.file_context:
        st.sidebar.success("✅ Document loaded successfully")

# Load model
@st.cache_resource
def load_custom_model():
    model_name = "amiguel/unsloth_finetune_test"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    return pipeline("text-classification", model=model, tokenizer=tokenizer)

# Generate response
def generate_response(prompt):
    bio_triggers = ['who are you', 'ataliba', 'yourself', 'skilled at', 
                    'background', 'experience', 'valonylabs', 'totalenergies']
    
    if any(trigger in prompt.lower() for trigger in bio_triggers):
        for line in ATALIBA_BIO.split('\n'):
            yield line + '\n'
            time.sleep(0.1)
        return

    try:
        classifier = load_custom_model()
        result = classifier(prompt)[0]
        label = result['label']
        score = result['score']
        context = st.session_state.file_context or "No document loaded."
        
        response_text = f"\n📘 **Prediction**: `{label}`\n📊 **Confidence**: `{score:.2%}`\n🗂️ **Context**: `{context[:300]}...`"
        for line in response_text.split('\n'):
            yield line + '\n'
            time.sleep(0.1)

    except Exception as e:
        yield f"⚠️ Model Error: {str(e)}"

# Chat interface
for msg in st.session_state.chat_history:
    with st.chat_message(msg["role"], avatar=USER_AVATAR if msg["role"] == "user" else BOT_AVATAR):
        st.markdown(msg["content"])

if prompt := st.chat_input("Ask about documents or technical matters..."):
    st.session_state.chat_history.append({"role": "user", "content": prompt})
    with st.chat_message("user", avatar=USER_AVATAR):
        st.markdown(prompt)

    with st.chat_message("assistant", avatar=BOT_AVATAR):
        response_placeholder = st.empty()
        full_response = ""

        for chunk in generate_response(prompt):
            full_response += chunk
            response_placeholder.markdown(full_response + "▌")

        response_placeholder.markdown(full_response)
        st.session_state.chat_history.append({"role": "assistant", "content": full_response})