import streamlit as st from transformers import AutoTokenizer, AutoModelForSeq2SeqLM from youtubesearchpython import VideosSearch # Load FLAN-T5 st.set_page_config(page_title="🔧 AI Vehicle Repair Assistant", layout="centered") @st.cache_resource def load_model(): tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-base") model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-base") return tokenizer, model tokenizer, model = load_model() # App UI st.title("🔧 AI Vehicle Repair Assistant") st.caption("Get brief repair instructions and a helpful YouTube video.") # Inputs vehicle = st.selectbox("Choose Vehicle Type:", ["Car", "Bike"]) issue = st.text_area("Describe the issue (e.g. engine won't start, AC not cooling):", height=100) # Button if st.button("🔍 Get Repair Instructions"): if issue.strip() == "": st.warning("Please enter a vehicle issue.") else: # Prompt for FLAN-T5 prompt = f"Give a short repair tip for a {vehicle.lower()} with this issue: {issue}." inputs = tokenizer(prompt, return_tensors="pt", max_length=128, truncation=True) outputs = model.generate(**inputs, max_new_tokens=100) instruction = tokenizer.decode(outputs[0], skip_special_tokens=True) # Show AI output st.markdown("### 🛠 Repair Tip:") st.success(instruction) # Search YouTube query = f"{vehicle} {issue} repair" videos_search = VideosSearch(query, limit=1) results = videos_search.result()["result"] if results: video_url = results[0]["link"] st.markdown("### 📺 Watch Tutorial:") st.video(video_url) else: st.info("No matching video found.")