Update main.py

main.py

@@ -70,7 +70,6 @@ app = FastAPI(
     description="Human-like AI Assistant for Harshal's Portfolio"
 )
 
-# Allow everything (your Next.js frontend)
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -80,7 +79,7 @@ app.add_middleware(
 )
 
 # ============================================================
-# 1) LOAD MAIN CHAT MODEL (Phi-3 Mini — CPU Friendly)
+# 1) LOAD MAIN CHAT MODEL
 # ============================================================
 
 LLM_MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"
@@ -96,66 +95,45 @@ llm_model.eval()
 print("✅ LLM Loaded Successfully")
 
 # ============================================================
-# 2) LOAD EMBEDDING MODEL + BUILD RAG FROM resume.pdf
+# 2) RAG: LOAD EMBEDDINGS FROM RESUME
 # ============================================================
 
 EMBED_MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
 RESUME_PATH = "resume.pdf"
 
-print(f"🔎 Loading embedding model: {EMBED_MODEL_NAME}")
 embedder = SentenceTransformer(EMBED_MODEL_NAME)
-print("✅ Embedding model loaded.")
+resume_vectors = None
 
-resume_vectors = None  # {"chunks": [...], "embeddings": tensor[]}
 
-
-def chunk_text(text: str, max_chars=450, overlap=80):
+def chunk_text(text, max_chars=450, overlap=80):
     text = " ".join(text.replace("\n", " ").split())
-    chunks = []
-    start = 0
-
+    chunks, start = [], 0
     while start < len(text):
         end = start + max_chars
         chunks.append(text[start:end])
         start = end - overlap
-
     return chunks
 
 
 def build_resume_rag():
-    """Reads resume.pdf, chunks it, builds embeddings"""
     global resume_vectors
-
     if not os.path.exists(RESUME_PATH):
         print("⚠️ resume.pdf missing — RAG disabled.")
         return
 
-    print("📄 Reading resume.pdf...")
     try:
         reader = PdfReader(RESUME_PATH)
-        text = ""
-
-        for page in reader.pages:
-            text += page.extract_text() or ""
-
+        text = "".join((p.extract_text() or "") for p in reader.pages)
         chunks = chunk_text(text)
-        print(f"✂️ Resume split into {len(chunks)} chunks.")
 
         embeddings = embedder.encode(
-            chunks,
-            convert_to_tensor=True,
-            normalize_embeddings=True
+            chunks, convert_to_tensor=True, normalize_embeddings=True
         )
 
-        resume_vectors = {
-            "chunks": chunks,
-            "embeddings": embeddings
-        }
-
+        resume_vectors = {"chunks": chunks, "embeddings": embeddings}
         print("✅ Resume RAG ready.")
-
     except Exception as e:
-        print(f"❌ Failed to build RAG: {e}")
+        print(f"❌ RAG Error: {e}")
         resume_vectors = None
 
 
@@ -166,29 +144,29 @@ def embed_query(query: str):
 def retrieve_relevant_resume_text(query: str, top_k=3):
     if resume_vectors is None:
         return ""
-
     try:
         q = embed_query(query)
         sims = torch.nn.functional.cosine_similarity(q.unsqueeze(0), resume_vectors["embeddings"])
-        top = torch.topk(sims, k=min(top_k, len(sims)))
-        return "\n\n".join(resume_vectors["chunks"][i] for i in top.indices)
+        top = torch.topk(sims, k=min(top_k, len(sims))).indices
+        return "\n\n".join(resume_vectors["chunks"][i] for i in top)
     except:
         return ""
 
 
-# Build RAG on startup
 build_resume_rag()
 
 # ============================================================
-# 3) CLEAN REQUEST SCHEMA FOR SWAGGER
+# 3) CLEAN REQUEST SCHEMA
 # ============================================================
 
 class Message(BaseModel):
     role: str
     content: str
 
+
 class ChatRequest(BaseModel):
-    messages: List<Message]
+    messages: List[Message]
+
 
 # ============================================================
 # 4) CHAT ENDPOINT
@@ -196,44 +174,28 @@ class ChatRequest(BaseModel):
 
 @app.post("/chat")
 def chat(req: ChatRequest):
-    """
-    Chat endpoint for Harshal AI.
-    Expects:
-    {
-        "messages": [
-            { "role": "user", "content": "Where are you from?" }
-        ]
-    }
-    Returns:
-    { "reply": "..." }
-    """
-
     try:
         if not req.messages:
             return {"reply": "I didn't get any question yet 🙂"}
 
         user_msg = req.messages[-1].content.strip()
-
         if not user_msg:
             return {"reply": "Could you repeat that? I didn't catch it."}
 
-        # ---- GET RAG TEXT ----
         resume_context = retrieve_relevant_resume_text(user_msg)
 
-        # ---- Persona ----
         system_prompt = f"""
-You are Harshal Sonawane, a real human software engineer from Pune, India.
-You speak casually, friendly, confident.
-You NEVER say you are an AI or model.
-You answer in short 2–4 sentences.
-Use resume details if helpful.
-If you don't know something, answer honestly.
-
-Relevant resume details:
+You are Harshal Sonawane, a real human software engineer from Pune.
+Speak casually, friendly, confident.
+Never say you are an AI.
+Keep replies short (2–4 sentences).
+Use resume context when helpful.
+If unknown, answer honestly.
+
+Resume info:
 {resume_context}
 """.strip()
 
-        # ---- Chat format for Phi-3 ----
         chat_messages = [
             {"role": "system", "content": system_prompt},
             {"role": "user", "content": user_msg},
@@ -260,13 +222,14 @@ Relevant resume details:
         reply = tokenizer.decode(generated, skip_special_tokens=True).strip()
 
         if not reply:
-            reply = "Sorry, something went wrong. Could you ask again?"
+            reply = "Something went wrong. Try again?"
 
         return {"reply": reply}
 
     except Exception as e:
         print(f"❌ Chat Error: {e}")
-        return {"reply": "Oops, I hit a snag. Please try again shortly."}
+        return {"reply": "Oops, I hit a snag. Try again shortly."}
+
 
 # ============================================================
 # 5) ROOT HEALTH CHECK
@@ -275,3 +238,4 @@ Relevant resume details:
 @app.get("/")
 def home():
     return {"message": "Harshal AI backend running with Phi-3 Mini + RAG 🎯"}
+