Upload 3 files

Dockerfile.txt

@@ -0,0 +1,20 @@
+# Gunakan image dasar Python 3.9
+FROM python:3.9-slim
+
+# Tetapkan direktori kerja di dalam container
+WORKDIR /app
+
+# Salin file requirements dulu untuk caching yang lebih efisien
+COPY requirements.txt .
+
+# Install semua library Python yang dibutuhkan
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Salin semua sisa kode backend ke dalam container
+COPY . .
+
+# Beri tahu Docker port mana yang akan diekspos oleh aplikasi kita
+EXPOSE 8000
+
+# Perintah untuk menjalankan server FastAPI saat container dimulai
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

main.py

@@ -0,0 +1,142 @@
+# main.py (Versi Final Definitif - Dengan Parsing Cerdas & Prompt Tepat)
+import os
+import shutil
+import re
+import uvicorn
+from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+import torch
+
+from langchain_community.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain_community.retrievers import BM25Retriever
+from langchain.retrievers import EnsembleRetriever
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch._dynamo as dynamo
+
+dynamo.config.automatic_dynamic_shapes = False
+dynamo.config.assume_static_by_default = True
+
+UPLOAD_DIR = "temp_uploads"
+os.makedirs(UPLOAD_DIR, exist_ok=True)
+
+app = FastAPI(title="Financial RAG Chatbot API")
+
+origins = ["*"]
+app.add_middleware(CORSMiddleware, allow_origins=origins, allow_credentials=True, allow_methods=["*"], allow_headers=["*"])
+
+rag_pipeline = {"retriever": None, "llm": None, "tokenizer": None, "embeddings": None, "all_chunks": None}
+
+def setup_rag_pipeline():
+    print("Memulai setup RAG pipeline...")
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    print(f"Menggunakan device: {device}")
+    print("Memuat Embedding Model...")
+    rag_pipeline["embeddings"] = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={'device': device})
+    print("Memuat LLM (Gemma 270M)...")
+    rag_pipeline["tokenizer"] = AutoTokenizer.from_pretrained("google/gemma-3-270m-it")
+    rag_pipeline["llm"] = AutoModelForCausalLM.from_pretrained("google/gemma-3-270m-it", device_map="auto", torch_dtype=torch.bfloat16)
+    print("Setup RAG pipeline selesai.")
+
+@app.on_event("startup")
+async def startup_event():
+    setup_rag_pipeline()
+
+@app.post("/upload")
+async def upload_document(file: UploadFile = File(...)):
+    # Fungsi ini tetap sama
+    try:
+        file_path = os.path.join(UPLOAD_DIR, file.filename)
+        with open(file_path, "wb") as buffer: shutil.copyfileobj(file.file, buffer)
+        loader = PyPDFLoader(file_path)
+        docs = loader.load()
+        text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=200)
+        chunks = text_splitter.split_documents(docs)
+        rag_pipeline["all_chunks"] = chunks
+        faiss_db = FAISS.from_documents(chunks, rag_pipeline["embeddings"])
+        faiss_retriever = faiss_db.as_retriever(search_kwargs={"k": 10})
+        bm25_retriever = BM25Retriever.from_documents(chunks)
+        bm25_retriever.k = 10
+        rag_pipeline["retriever"] = EnsembleRetriever(retrievers=[bm25_retriever, faiss_retriever], weights=[0.5, 0.5])
+        return {"status": "sukses", "message": f"Dokumen '{file.filename}' berhasil diproses."}
+    except Exception as e: raise HTTPException(status_code=500, detail=str(e))
+    finally:
+        if 'file_path' in locals() and os.path.exists(file_path): os.remove(file_path)
+
+class ChatRequest(BaseModel):
+    query: str
+
+@app.post("/chat")
+async def chat_with_doc(request: ChatRequest):
+    if not rag_pipeline.get("retriever"):
+        raise HTTPException(status_code=400, detail="Dokumen belum di-upload.")
+
+    query_original = request.query
+    query_lower = query_original.lower()
+    clean_context = ""
+    found_source = ""
+    
+    priority_keywords = ["jumlah aset lancar"]
+    use_smart_lane = any(keyword in query_lower for keyword in priority_keywords)
+
+    if use_smart_lane:
+        print("Jalur Cerdas Aktif! Mencari konteks & parsing tahun...")
+        year_match = re.search(r'\b(202[3-4])\b', query_lower)
+        target_year = year_match.group(1) if year_match else "2024"
+        
+        all_chunks = rag_pipeline.get("all_chunks", [])
+        for chunk in all_chunks:
+            lines = chunk.page_content.split('\n')
+            for line in lines:
+                if any(keyword in line.lower() for keyword in priority_keywords):
+                    # --- [FIX #1] PARSING CERDAS DENGAN REGEX ---
+                    # Mencari semua angka yang diformat dengan koma/titik
+                    numbers = re.findall(r'(\d{1,3}(?:[.,]\d{3})*)', line)
+                    if len(numbers) >= 2:
+                        try:
+                            # Asumsi angka pertama adalah 2024, kedua 2023
+                            value_2024 = numbers[0]
+                            value_2023 = numbers[1]
+                            value = value_2024 if target_year == "2024" else value_2023
+                            # Buat fakta yang bersih untuk LLM
+                            clean_context = f"Fakta: jumlah aset lancar untuk tahun {target_year} adalah {value}."
+                            found_source = f"Halaman {chunk.metadata.get('page', 'NA')}"
+                            break
+                        except (IndexError, ValueError):
+                            continue
+            if clean_context:
+                break
+    
+    if not clean_context:
+        print("Menggunakan Jalur Normal (Hybrid Search)...")
+        retrieved_docs = rag_pipeline["retriever"].invoke(query_original)
+        clean_context = "\n\n".join([doc.page_content for doc in retrieved_docs[:3]])
+        found_source = ", ".join(list(set([f"Halaman {doc.metadata.get('page', 'NA')}" for doc in retrieved_docs[:3]])))
+
+    # --- Generation ---
+    tokenizer = rag_pipeline["tokenizer"]
+    model = rag_pipeline["llm"]
+    
+    # --- [FIX #2] PROMPT YANG SANGAT TO-THE-POINT ---
+    chat_template = [
+        {"role": "user", "content": f"Gunakan informasi ini: '{clean_context}'. Jawab pertanyaan ini: '{query_original}'"}
+    ]
+    final_prompt = tokenizer.apply_chat_template(chat_template, tokenize=False, add_generation_prompt=True)
+    inputs = tokenizer(final_prompt, return_tensors="pt").to(model.device)
+    outputs = model.generate(**inputs, max_new_tokens=150, do_sample=False, pad_token_id=tokenizer.eos_token_id)
+    
+    input_length = inputs.input_ids.shape[1]
+    generated_tokens = outputs[0, input_length:]
+    final_answer = tokenizer.decode(generated_tokens, skip_special_tokens=True)
+
+    return {"answer": final_answer.strip(), "source": found_source}
+
+@app.get("/")
+def read_root():
+    return {"message": "Selamat datang di Financial RAG Chatbot API!"}
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=8000)

requirements.txt

@@ -0,0 +1,13 @@
+fastapi
+uvicorn[standard]
+python-multipart
+torch
+transformers
+accelerate
+bitsandbytes
+langchain
+sentence-transformers
+pypdf
+faiss-cpu
+rank_bm25
+langchain_community