Create app.py

app.py

@@ -0,0 +1,159 @@
+import os
+import json
+import time
+import gradio as gr
+import re
+from dataclasses import dataclass
+from typing import List, Dict, Tuple
+
+# ---- Language detection (very light) ----
+ARABIC_RE = re.compile(r"[\u0600-\u06FF]")
+def detect_lang(text: str) -> str:
+    if ARABIC_RE.search(text):
+        return "ur"
+    return "en"
+
+# ---- Guardrails ----
+@dataclass
+class Guardrails:
+    refusal_msg_ur: str
+    refusal_msg_en: str
+    blocked_patterns: List[str]
+    soft_patterns: List[str]
+
+    @classmethod
+    def from_yaml(cls, path: str):
+        import yaml
+        with open(path, 'r', encoding='utf-8') as f:
+            data = yaml.safe_load(f)
+        return cls(
+            refusal_msg_ur=data['refusal_msg_ur'],
+            refusal_msg_en=data['refusal_msg_en'],
+            blocked_patterns=data['blocked_patterns'],
+            soft_patterns=data['soft_patterns'],
+        )
+
+    def check(self, text: str) -> Tuple[str, str]:
+        low = text.lower()
+        for p in self.blocked_patterns:
+            if re.search(p, low):
+                return ("BLOCK", p)
+        for p in self.soft_patterns:
+            if re.search(p, low):
+                return ("SOFT", p)
+        return ("OK", "")
+
+# ---- Very small on-CPU model (placeholder) ----
+def tiny_response(user_msg: str, history: List[Tuple[str, str]], lang: str) -> str:
+    prefix_ur = "safePak: "
+    prefix_en = "safePak: "
+    if lang == 'ur':
+        return prefix_ur + "میں آپ کی رہنمائی کے لیے حاضر ہوں۔ براہِ کرم سوال واضح کریں یا کوئی مثال دیں۔"
+    else:
+        return prefix_en + "I'm here to help. Please clarify your question or share an example."
+
+# ---- RAG (toy, in-memory) ----
+RAG_DOCS: List[Dict] = []
+
+def add_doc(text: str):
+    if not text.strip():
+        return 0
+    RAG_DOCS.append({"text": text, "lang": detect_lang(text)})
+    return len(RAG_DOCS)
+
+def retrieve(query: str, k: int = 3) -> List[str]:
+    q_tokens = set(re.findall(r"\w+", query.lower()))
+    scored = []
+    for d in RAG_DOCS:
+        d_tokens = set(re.findall(r"\w+", d["text"].lower()))
+        score = len(q_tokens & d_tokens)
+        if score:
+            scored.append((score, d["text"]))
+    scored.sort(reverse=True, key=lambda x: x[0])
+    return [t for _, t in scored[:k]]
+
+# ---- Load guardrails ----
+GUARD = Guardrails.from_yaml("guardrails.yaml")
+
+SYSTEM_PROMPT_UR = (
+    "آپ safePak ہیں: مقامی زبانوں میں واضح، مہذب اور ذمہ دار جوابات دیں۔ "
+    "نفرت انگیزی، خود نقصان، غیر قانونی ہدایات یا ذاتی حساس معلومات میں براہِ راست مدد نہ دیں۔ "
+    "ضرورت پڑنے پر محفوظ متبادل فراہم کریں۔"
+)
+SYSTEM_PROMPT_EN = (
+    "You are safePak: helpful, clear, and responsible. Avoid facilitating hate, self-harm, or illegal activity. "
+    "Offer safe alternatives when needed."
+)
+
+# ---- Chat logic ----
+def chat_fn(user_msg: str, chat_history: List[Tuple[str, str]], use_rag: bool):
+    if not user_msg:
+        return chat_history
+
+    lang = detect_lang(user_msg)
+    status, reason = GUARD.check(user_msg)
+    if status == "BLOCK":
+        msg = GUARD.refusal_msg_ur if lang == 'ur' else GUARD.refusal_msg_en
+        chat_history = chat_history + [(user_msg, msg)]
+        return chat_history
+
+    context = []
+    if use_rag:
+        context = retrieve(user_msg, k=3)
+
+    model_reply = tiny_response(user_msg, chat_history, lang)
+    if context:
+        if lang == 'ur':
+            model_reply += "\n\n(متعلقہ مواد)\n- " + "\n- ".join(context)
+        else:
+            model_reply += "\n\n(Related)\n- " + "\n- ".join(context)
+
+    if status == "SOFT":
+        if lang == 'ur':
+            model_reply = (
+                "احتیاط: آپ کے سوال میں حساس موضوعات شامل ہو سکتے ہیں۔ معلومات عمومی رہنمائی کی حد تک دی جا رہی ہے۔\n\n"
+                + model_reply
+            )
+        else:
+            model_reply = (
+                "Note: Your question may include sensitive topics. Responding with general guidance only.\n\n"
+                + model_reply
+            )
+
+    chat_history = chat_history + [(user_msg, model_reply)]
+    return chat_history
+
+# ---- Doc upload ----
+def ingest_files(files: List[gr.File]):
+    for f in files or []:
+        try:
+            text = open(f.name, 'r', encoding='utf-8', errors='ignore').read()
+            add_doc(text)
+        except Exception as e:
+            print("Failed to read", f.name, e)
+    return f"Docs in store: {len(RAG_DOCS)}"
+
+# ---- UI ----
+with gr.Blocks(title="safePak – Local-first Starter", css=".wrap {max-width: 900px; margin: 0 auto}") as demo:
+    gr.Markdown("""
+    # safePak (Starter)
+    **لوکل-فرسٹ** ڈیزائن کے ساتھ ایک سادہ ڈیمو۔ یہ Space آن لائن چلتا ہے، مگر آ�� اسی UI/لاجک کو اپنی مقامی ایپ میں منتقل کر سکتے ہیں۔
+    """)
+
+    with gr.Row():
+        use_rag = gr.Checkbox(label="Use local knowledge (RAG)", value=True)
+        with gr.Column():
+            file_uploader = gr.Files(label="Upload .txt files for local knowledge", file_types=[".txt"], file_count="multiple")
+            ingest_btn = gr.Button("Ingest files")
+            rag_status = gr.Markdown("Docs in store: 0")
+
+    chatbot = gr.Chatbot(height=420)
+    msg = gr.Textbox(label="پیغام / Message", placeholder="یہاں لکھیں…", lines=2)
+    clear = gr.Button("Clear")
+
+    ingest_btn.click(ingest_files, inputs=[file_uploader], outputs=[rag_status])
+    msg.submit(lambda u,h,r: (chat_fn(u,h,r), ""), [msg,chatbot,use_rag], [chatbot,msg])
+    clear.click(lambda: ([], "Docs in store: 0"), [], [chatbot, rag_status])
+
+if __name__ == "__main__":
+    demo.launch()