Update api.py

api.py

@@ -7,44 +7,47 @@ from langdetect import detect
 from deep_translator import GoogleTranslator
 from sentence_transformers import SentenceTransformer
 from pinecone import Pinecone
+from openai import OpenAI
 
 # ---------- Config ----------
 DATASET_PATH = "data/coaching_millionaer_dataset.json"
 load_dotenv(override=True)
 
-OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") or os.environ.get("OPENAI_API_KEY")
-PINECONE_API_KEY = os.getenv("PINECONE_API_KEY") or os.environ.get("PINECONE_API_KEY")
+# Load secrets from Hugging Face Space
+HF_TOKEN = os.getenv("HF_TOKEN")
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
 PINECONE_INDEX_NAME = "ebook"
 
 # ---------- App ----------
 app = Flask(__name__)
 CORS(app, resources={r"/ask": {"origins": "*"}})
 
-# ---------- OpenAI Client ----------
-from openai import OpenAI
-import inspect
-
+# ---------- LLM Client ----------
 client = None
 try:
-    if not OPENAI_API_KEY or OPENAI_API_KEY.strip() == "":
-        raise ValueError("Missing OPENAI_API_KEY in environment variables")
-
-    # Handle Hugging Face’s automatic proxy injection safely
-    init_params = {"api_key": OPENAI_API_KEY}
-    if "proxies" in inspect.signature(OpenAI.__init__).parameters:
-        init_params["proxies"] = None
-
-    client = OpenAI(**init_params)
-    print(f"✅ Using OpenAI SDK version {client.__module__.split('.')[0]} (modern client)")
+    if HF_TOKEN:
+        # Use Hugging Face Inference Provider
+        client = OpenAI(
+            base_url="https://router.huggingface.co/v1",
+            api_key=HF_TOKEN,
+        )
+        print("✅ Using Hugging Face Inference Provider (OpenAI-compatible API)")
+    elif OPENAI_API_KEY:
+        # Fallback to OpenAI if provided
+        client = OpenAI(api_key=OPENAI_API_KEY)
+        print("✅ Using OpenAI client directly")
+    else:
+        raise ValueError("No valid API key found. Set HF_TOKEN or OPENAI_API_KEY.")
 except Exception as e:
+    print(f"❌ Failed to initialize LLM client: {e}")
     client = None
-    print(f"❌ Failed to initialize OpenAI client: {e}")
 
 # ---------- Retriever ----------
 retriever = None
 try:
     if not PINECONE_API_KEY:
-        raise ValueError("PINECONE_API_KEY missing in .env")
+        raise ValueError("PINECONE_API_KEY missing in environment variables")
 
     pc = Pinecone(api_key=PINECONE_API_KEY)
     index = pc.Index(PINECONE_INDEX_NAME)
@@ -105,14 +108,14 @@ def system_prompt_book_only() -> str:
         "If the user asks about people like Javid Niazi-Hoffmann, describe them factually using the book content. "
         "Mention page numbers where possible. "
         "If the context is not relevant, say you don’t have that information in the book and provide a general, helpful answer. "
-        "Always respond in the same language as the user's question, even if the book content is in another language."
+        "Always respond in the same language as the user's question."
     )
 
 def system_prompt_fallback() -> str:
     return (
         "You are CoachingBot, a helpful business and life mentor. "
         "The question cannot be answered from the book, so answer using your general coaching knowledge. "
-        "Always respond in the same language as the user's question, even if the book content is in another language. "
+        "Always respond in the same language as the user's question. "
         "Do not invent book citations."
     )
 
@@ -138,7 +141,7 @@ def health():
         {
             "status": "running",
             "retriever_ready": bool(retriever),
-            "openai_key_loaded": bool(OPENAI_API_KEY),
+            "hf_key_loaded": bool(HF_TOKEN),
             "pinecone_key_loaded": bool(PINECONE_API_KEY),
             "index_name": PINECONE_INDEX_NAME,
         }
@@ -185,13 +188,13 @@ def ask():
         sys_prompt = system_prompt_fallback()
         user_content = question
 
-    # ---------- OpenAI Query ----------
     if client is None:
-        return jsonify(format_answers(question, "⚠️ OpenAI client not initialized.", results)), 200
+        return jsonify(format_answers(question, "⚠️ No language model initialized.", results)), 200
 
+    # ---------- LLM Query ----------
     try:
         response = client.chat.completions.create(
-            model="gpt-4o-mini",
+            model="openai/gpt-oss-120b:cerebras",  # Hugging Face model
             messages=[
                 {"role": "system", "content": sys_prompt},
                 {"role": "user", "content": user_content},
@@ -201,7 +204,7 @@ def ask():
         answer = response.choices[0].message.content.strip()
     except Exception as e:
         traceback.print_exc()
-        return jsonify(format_answers(question, f"⚠️ OpenAI call failed: {e}", results)), 200
+        return jsonify(format_answers(question, f"⚠️ LLM call failed: {e}", results)), 200
 
     return jsonify(format_answers(question, answer, results))