Create app.py

app.py

@@ -0,0 +1,48 @@
+import gradio as gr
+from chromadb_semantic_search_for_dataset import search_cases
+from transformers import pipeline
+
+# Load small multilingual LLM
+qa_pipeline = pipeline(
+    "text2text-generation",
+    model="google/flan-t5-small"  # Change to a better Nepali-supporting model if needed
+)
+
+def semantic_search_only(search_text):
+    results, _ = search_cases(search_text)
+    return results
+
+def rag_answer(question, search_text):
+    # First get top 5 cases
+    _, context = search_cases(search_text)
+    
+    # Build RAG prompt
+    prompt = (
+        f"तपाईं एक कानूनी सहायक हुनुहुन्छ। तलका केसहरूको जानकारी प्रयोग गरेर प्रश्नको जवाफ दिनुहोस्।\n\n"
+        f"सन्दर्भ:\n{context}\n\n"
+        f"प्रश्न: {question}\n"
+        f"उत्तर:"
+    )
+    
+    # Generate answer
+    answer = qa_pipeline(prompt, max_length=512)[0]['generated_text']
+    return answer
+
+# Gradio UI with two sections
+with gr.Blocks() as iface:
+    gr.Markdown("# 📚 Semantic Search + RAG for Nepali Legal Cases")
+
+    with gr.Tab("🔍 Semantic Search"):
+        search_box = gr.Textbox(label="Search for a case")
+        search_output = gr.Textbox(label="Top 5 Similar Cases")
+        search_button = gr.Button("Search")
+        search_button.click(fn=semantic_search_only, inputs=search_box, outputs=search_output)
+
+    with gr.Tab("🤖 Ask a Question (RAG)"):
+        rag_query = gr.Textbox(label="Ask your question")
+        rag_search_context = gr.Textbox(label="Search for a case (context)")
+        rag_output = gr.Textbox(label="LLM Answer")
+        rag_button = gr.Button("Get Answer")
+        rag_button.click(fn=rag_answer, inputs=[rag_query, rag_search_context], outputs=rag_output)
+
+iface.launch()