Update app.py

app.py

@@ -1,63 +1,137 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+import pickle
+import faiss
+import numpy as np
+import torch
+import os
+from transformers import AutoTokenizer, AutoModel
+from openai import OpenAI
+from dotenv import load_dotenv
 
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+load_dotenv()
+api = os.getenv("OPENAI_API_KEY")
+client = OpenAI(api_key=api)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# Load IndoLegalBERT
+tokenizer = AutoTokenizer.from_pretrained("archi-ai/Indo-LegalBERT")
+model = AutoModel.from_pretrained("archi-ai/Indo-LegalBERT")
 
-    messages.append({"role": "user", "content": message})
+# Pooling dengan mean pooling
+def get_embedding(text):
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512, padding="max_length")
+    with torch.no_grad():
+        outputs = model(**inputs)
+        last_hidden = outputs.last_hidden_state
+        mask = inputs["attention_mask"].unsqueeze(-1).expand(last_hidden.size()).float()
+        masked = last_hidden * mask
+        summed = torch.sum(masked, 1)
+        counts = torch.clamp(mask.sum(1), min=1e-9)
+        mean_pooled = summed / counts
+        return mean_pooled.squeeze().numpy()
 
-    response = ""
+# Generate all embeddings
+embeddings = np.array([get_embedding(text) for text in texts])
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
 
-        response += token
-        yield response
+# 5. Simpan ke FAISS
+dimension = embeddings.shape[1]
+index = faiss.IndexFlatL2(dimension)
+index.add(embeddings)
 
+# 6. Simpan FAISS index dan metadata
+faiss.write_index(index, "legal_index.faiss")
+with open("legal_metadata.pkl", "wb") as f:
+    pickle.dump(titles, f)
 
+# 2. Load FAISS index dan metadata
+index = faiss.read_index("legal_index.faiss")
+with open("legal_metadata.pkl", "rb") as f:
+    metadata = pickle.load(f)
+
+# 4. Fungsi pencarian pasal hukum terkait
+def search_laws(query, top_k=3):
+    vec = get_embedding(query).reshape(1, -1)
+    D, I = index.search(vec, top_k)
+    results = []
+    for i in I[0]:
+        if i < len(metadata):
+            results.append(f"- {metadata[i]}\n{texts[i]}")
+    return results
+
+# 5. Fungsi untuk membentuk prompt ke OpenAI
+def build_prompt(query, contexts):
+    context_text = "\n\n".join(contexts)
+    return f"""
+Anda adalah asisten hukum berbasis hukum Indonesia.
+
+Permintaan pengguna:
+\"{query}\"
+
+Gunakan konteks hukum berikut:
+{context_text}
+
+Berikan penjelasan hukum yang sistematis dan profesional. Sebutkan pasal hukum jika ada.
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+
+# 6. Fungsi untuk interaksi LLM (pakai GPT-3.5 Turbo)
+openai.api_key = "YOUR_OPENAI_API_KEY"  # <- Ganti dengan API key milikmu
+
+def ask_llm(query):
+    contexts = search_laws(query)
+    prompt = build_prompt(query, contexts)
+
+    response = client.chat.completions.create(
+        model="gpt-3.5-turbo",
+        messages=[
+            {"role": "system", "content": "Anda adalah ahli hukum Indonesia."},
+            {"role": "user", "content": prompt}
+        ],
+        temperature=0.2,
+       # max_tokens=512,
+    )
+    return response.choices[0].message.content
+
+# Gradio UI
+# Fungsi simulasi RAG Legal Agent
+def rag_legal_analysis(document_text, issue_type):
+    if issue_type == "Analisis Syarat Sah Perjanjian":
+        return ask_llm(document_text)
+    elif issue_type == "Deteksi Klausul Bermasalah":
+        return ask_llm(document_text)
+    elif issue_type == "Risiko Hukum Pihak Tertentu":
+        return ask_llm(document_text)
+    else:
+        return "Silakan pilih jenis analisis hukum yang ingin dilakukan."
+
+# Gradio UI
+with gr.Blocks(title="Naraya Smart Legal Assitant") as demo:
+    gr.Markdown("# 🤖 Naraya Smart Legal Assitant")
+    gr.Markdown("Masukkan isi perjanjian atau kontrak, lalu pilih jenis analisis hukum.")
+
+    document_input = gr.Textbox(
+                        label="Isi Dokumen Kontrak", 
+                        lines=10, 
+                        placeholder="Masukkan isi kontrak di sini atau upload dokumen")
+        #document_input = gr.MultimodalTextbox(
+         #               interactive=True, 
+          #              label="Isi Dokumen Kontrak", 
+           #             lines=10, 
+            #            placeholder="Masukkan isi kontrak di sini atau upload dokumen")
+    issue_type = gr.Radio(
+        label="Jenis Analisis Hukum",
+        choices=[
+            "Analisis Syarat Sah Perjanjian",
+            "Deteksi Klausul Bermasalah",
+            "Risiko Hukum Pihak Tertentu"
+        ]
+    )
+    output = gr.Textbox(label="Hasil Analisis Hukum", lines=20)
+
+    analyze_button = gr.Button("🔍 Analisa Sekarang")
+    analyze_button.click(fn=rag_legal_analysis, inputs=[document_input, issue_type], outputs=output)    
 
 
 if __name__ == "__main__":