At least 20 e-commerce businesses in Indonesia have stopped operating due to human resource problems. The recommendation system is an important feature in e-commerce which was initially done manually. However, a computing field has emerged that can replace human labor in recommendation systems, reduce human errors and work more efficiently. This paper aims to implement an efficient and scalable recommendation system using Machine Learning techniques. The content-based filtering used in this paper uses the Approximate Nearest Neighbor algorithm with different indexes that provide similar product recommendations. Data is collected from Tokopedia product information, which is divided into five categories. The optimal Nprobe is found at 6% which is the standard Nprobe for each index. The index that has the best recall@R parameter, fastest prediction time and training time respectively are IDMap,Flat, IVF1024_HNSW32,Flat and IDMap,Flat.