Update core/rag_services.py

core/rag_services.py

@@ -1,12 +1,11 @@
 # ==========================================================
-# File: core/rag_services.py (FINAL STABLE – Okt 2025)
+# File: core/rag_services.py (FINAL PATCH - Okt 2025)
 # ----------------------------------------------------------
 # ✅ Bisa baca PDF, DOCX, TXT
-# ✅ Aman di Hugging Face (gunakan /tmp)
 # ✅ Incremental indexing + manifest
 # ✅ Auto rebuild index
-# ✅ Fix "'dict' object has no attribute 'replace'"
-# ✅ Fix "'ChatPromptValue' object is not subscriptable"
+# ✅ Fix 'replace' & 'ChatPromptValue' errors
+# ✅ Pipeline manual (tidak bergantung internal LangChain)
 # ==========================================================
 
 import os
@@ -43,8 +42,6 @@ except ImportError:
     from langchain_community.document_loaders import UnstructuredFileLoader as UnstructuredLoader
 
 from langchain_core.prompts import ChatPromptTemplate
-from langchain_core.output_parsers import StrOutputParser
-from langchain_core.runnables import RunnablePassthrough
 
 # ===============================
 # Path & Config
@@ -234,7 +231,7 @@ def incremental_index_update(embedding_model):
     return db
 
 # ===============================
-# Build RAG Chain (Final Fixed)
+# Build RAG Chain (Manual Pipeline)
 # ===============================
 def get_rag_chain(llm, vector_store):
     if vector_store is None or vector_store.index.ntotal == 0:
@@ -254,30 +251,33 @@ Pertanyaan:
 Jawaban:
 """)
 
-    def _combine_context(input_data):
-        q = input_data.get("question") if isinstance(input_data, dict) else str(input_data)
-        try:
-            docs = retriever.invoke(q)
-            return "\n\n".join([getattr(d, "page_content", "") for d in docs if d])
-        except Exception as e:
-            return f"(Gagal mengambil konteks: {e})"
-
-    def _extract_question(input_data):
-        if isinstance(input_data, dict):
-            return str(input_data.get("question", ""))
-        return str(input_data)
-
-    chain = (
-        {
-            "context": _combine_context,
-            "question": _extract_question,
-        }
-        | prompt
-        | llm
-        | StrOutputParser()
-    )
-
-    return chain
+    class RAGPipeline:
+        def __init__(self, retriever, prompt, llm):
+            self.retriever = retriever
+            self.prompt = prompt
+            self.llm = llm
+
+        def invoke(self, query):
+            # Normalisasi input
+            question = str(query.get("question", "")) if isinstance(query, dict) else str(query)
+
+            try:
+                docs = self.retriever.invoke(question)
+                context = "\n\n".join([getattr(d, "page_content", "") for d in docs if d])
+            except Exception as e:
+                context = f"(Gagal mengambil konteks: {e})"
+
+            formatted_prompt = self.prompt.format(context=context, question=question)
+
+            try:
+                output = self.llm.invoke(formatted_prompt)
+                if hasattr(output, "content"):
+                    return output.content
+                return str(output)
+            except Exception as e:
+                return f"[RAG Error] Gagal memanggil LLM: {e}"
+
+    return RAGPipeline(retriever, prompt, llm)
 
 # ===============================
 # Auto Init di Startup