Update app.py

app.py

@@ -1,43 +1,48 @@
-from langchain_community.document_loaders import TextLoader
-from langchain.text_splitter import CharacterTextSplitter
-from langchain_community.embeddings import HuggingFaceEmbeddings
-from langchain_community.vectorstores import Chroma
-from langchain_community.llms import HuggingFaceHub
-from langchain.chains import RetrievalQA
+from fastapi import FastAPI
 import gradio as gr
-
-# 1. Load plain text file (Arabic)
-loader = TextLoader("knowledge.txt", encoding="utf-8")
-docs = loader.load()
-
-# 2. Split into small chunks
-text_splitter = CharacterTextSplitter(separator="\n", chunk_size=500, chunk_overlap=50)
-documents = text_splitter.split_documents(docs)
-
-# 3. Create multilingual embeddings
-embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
-
-# 4. Use Chroma vector store (instead of FAISS)
-db = Chroma.from_documents(documents, embeddings)
-retriever = db.as_retriever()
-
-# 5. Load LLM from Hugging Face (free)
-llm = HuggingFaceHub(
-    repo_id="tiiuae/falcon-7b-instruct",
-    model_kwargs={"temperature": 0.3, "max_new_tokens": 200}
+from sentence_transformers import SentenceTransformer, util
+import torch
+
+# Load Arabic sentence transformer model
+model = SentenceTransformer("CAMeL-Lab/bert-base-arabic-camelbert-ca")
+
+# Load and preprocess knowledge base
+def load_knowledge(file_path):
+    with open(file_path, "r", encoding="utf-8") as f:
+        content = f.read()
+    passages = [p.strip() for p in content.split("\n\n") if p.strip()]
+    embeddings = model.encode(passages, convert_to_tensor=True)
+    return passages, embeddings
+
+passages, passage_embeddings = load_knowledge("knowledge.txt")
+
+# Search function
+def answer_question_arabic(query):
+    query_embedding = model.encode(query, convert_to_tensor=True)
+    scores = util.cos_sim(query_embedding, passage_embeddings)[0]
+    top_idx = torch.argmax(scores).item()
+    best_score = scores[top_idx].item()
+
+    if best_score < 0.4:
+        return "عذرًا، لم أتمكن من العثور على إجابة مناسبة في قاعدة المعرفة."
+    return passages[top_idx]
+
+# Gradio UI
+demo = gr.Interface(
+    fn=answer_question_arabic,
+    inputs=gr.Textbox(label="اكتب سؤالك هنا", placeholder="ما هي أهداف التنمية المستدامة؟"),
+    outputs=gr.Textbox(label="إجابة"),
+    title="روبوت المعرفة - التنمية المستدامة",
+    description="أجب عن الأسئلة باللغة العربية بناءً على قاعدة معرفية من وزارة المالية حول التنمية المستدامة والموازنة التشاركية."
 )
 
-# 6. Create QA chain
-qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)
+# FastAPI app
+app = FastAPI()
 
-# 7. Gradio interface
-def answer_question(question):
-    return qa_chain.run(question)
+@app.get("/")
+def read_root():
+    return {"message": "مرحبا بك! انتقل إلى /gradio لبدء التفاعل."}
 
-gr.Interface(
-    fn=answer_question,
-    inputs=gr.Textbox(label="اكتب سؤالك هنا", lines=2),
-    outputs=gr.Textbox(label="الإجابة"),
-    title="شات بوت معرفي عربي",
-    description="أدخل سؤالك حول محتوى ملف المعرفة بالعربية.",
-).launch(share=True)
+@app.get("/gradio")
+def launch_gradio():
+    return gr.mount_gradio_app(app, demo, path="/gradio")