api update commit final

app/rag_app.py

@@ -1,96 +1,97 @@
-
 # app/rag_app.py
 import os
-from groq import Groq
 import httpx
+from openai import OpenAI
 from .embeddings import EmbeddingManager
 from .store import VectorStore
 
+
 class RAGApp:
     def __init__(self):
         self.embedder = None
         self.vectorstore = None
         self.client = None
-        
+
         try:
+            # Initialize embedder and FAISS vectorstore
             self.embedder = EmbeddingManager()
             self.vectorstore = VectorStore()
-            api_key = os.getenv("GROQ_API_KEY")
-            custom_http_client = httpx.Client()
-            self.client = Groq(api_key=api_key, http_client=custom_http_client)
-        except Exception as e:
-            print("RAGApp init error:", e)
-            # Optionally, re-raise to crash on startup.
-            # raise RuntimeError(f"RAGApp init failed: {e}")
 
-    # ...rest unchanged...
+            # Load Hugging Face token
+            
+            api_key = os.getenv("HF_TOKEN")
+            if not api_key:
+                raise ValueError("HF_TOKEN not found in environment variables")
 
+            # Use Hugging Face OpenAI-compatible router
+            self.client = OpenAI(
+                base_url="https://router.huggingface.co/v1",
+                api_key=api_key,
+                http_client=httpx.Client(timeout=60.0),
+            )
+
+            print("✅ RAGApp initialized successfully with Hugging Face router.")
 
+        except Exception as e:
+            print("❌ RAGApp init error:", e)
+            self.client = None
 
+    # 🧠 Add notes and store embeddings
     def add_notes(self, text):
-        chunks = [text[i:i+1000] for i in range(0, len(text), 800)]
+        chunks = [text[i:i + 1000] for i in range(0, len(text), 800)]
         embeddings = self.embedder.generate_embeddings(chunks)
         self.vectorstore.add_documents(chunks, embeddings)
         return len(chunks)
 
+    # 💬 Query the system
     def ask(self, query):
         try:
-            # 1️⃣ Generate embedding for query
+            if not self.client:
+                return "Error: API client not initialized."
+
+            # 1️⃣ Create embedding for query
             q_embed = self.embedder.generate_embeddings([query])[0]
 
-            # 2️⃣ Retrieve most relevant chunks from vector store
+            # 2️⃣ Retrieve similar chunks
             docs = self.vectorstore.retrieve_similar_docs(q_embed, top_k=3)
-            context = "\n\n".join(docs)
+            context = "\n\n".join(docs) if docs else "No context found in notes."
 
-            # 3️⃣ Prepare the system and user prompts
+            # 3️⃣ Prepare the conversation
             messages = [
                 {
                     "role": "system",
                     "content": (
-                        "You are a world-class engineering tutor specializing in Electronics, Embedded Systems, and Programming. "
-                        "Your teaching style dynamically adapts based on the student's question type.\n\n"
-                        "### 🧩 Behavior Rules:\n"
-                        "1️⃣ If the question is conceptual, explain step-by-step with analogies and real-world relevance.\n"
-                        "2️⃣ If the question involves code, analyze, fix, and explain why the fix works.\n"
-                        "3️⃣ If hardware-related, combine theory with hardware behavior and signals.\n"
-                        "4️⃣ If theory from uploaded notes, summarize and add context from real-world applications.\n\n"
-                        "### 🧠 Response Structure:\n"
-                        "1. Motivation / Why It Matters\n"
-                        "2. Concept Breakdown / Explanation\n"
-                        "3. Analogy\n"
-                        "4. Code or Example\n"
-                        "5. Practical Insight\n"
-                        "6. Common Mistakes + Tips\n\n"
-                        "### ✨ Style Guidelines:\n"
-                        "- Use bold keywords and emojis, and Markdown for structure.\n"
-                        "- Be friendly yet technically precise.\n"
-                        "- Never say 'as an AI model'.\n"
-                        "- If context from notes is relevant, integrate it smoothly.\n\n"
-                        "Your goal: help the student truly understand the concept."
+                        "You are a world-class engineering tutor specializing in Electronics, Embedded Systems, and Programming.\n"
+                        "Your responses must be clear, technically accurate, and engaging.\n\n"
+                        "### Behavior:\n"
+                        "1. If the question is conceptual → explain with clarity and real-world relevance.\n"
+                        "2. If it involves code → analyze, correct, and explain fixes.\n"
+                        "3. If hardware-related → explain theory + circuit/signal behavior.\n"
+                        "4. If theory-based from uploaded notes → connect with practical examples.\n\n"
+                        "### Output Style:\n"
+                        "- Use Markdown.\n"
+                        "- Highlight key terms with bold text.\n"
+                        "- Use emojis and structured headings for readability.\n"
+                        "- Avoid phrases like 'as an AI model'."
                     ),
                 },
                 {
                     "role": "user",
-                    "content": f"Context:\n{context}\n\nQuestion: {query}\nAnswer:",
+                    "content": f"Context:\n{context}\n\nQuestion: {query}\nAnswer clearly and in detail below:",
                 },
             ]
 
-            # 4️⃣ Call Groq API
+            # 4️⃣ Call Hugging Face model via router
             completion = self.client.chat.completions.create(
-                model="openai/gpt-oss-20b",
+                model="openai/gpt-oss-20b",  # ✅ correct model route
                 messages=messages,
-                temperature=0.3,
+                temperature=0.4,
                 max_tokens=800,
-                top_p=1
             )
 
-            # 5️⃣ Extract response
-            for chunk in completion:
-                key, value = chunk
-                if key == 'choices':
-                    return value[0].message.content.strip()
+            # 5️⃣ Return the model's response
+            return completion.choices[0].message.content
 
-            return "No valid response from model."
 
         except Exception as e:
             print("❌ Error in ask():", e)