The growing adoption of cloud computing, driven by its high scalability and virtualization, has encouraged individuals and enterprises to outsource encrypted data to cloud servers. However, the encryption disrupts data relationships, complicating retrieval tasks. While modern searchable encryption (SE) techniques, such as Fully Homomorphic Encryption (FHE) and Neural Cryptography, enable secure file retrieval, untrusted cloud servers may tamper with data, delete files, or exploit trapdoors to infer information about newly inserted files. Existing single-keyword search approaches, including traditional Bloom Filters, often suffer from low precision, leading to unnecessary consumption of bandwidth and computational resources. To address these challenges, this paper proposes a state-of-the-art Multi-Keyword Searchable Encryption (MKSE) scheme using Transformer-based Secure Embeddings (TSE). The scheme enhances search accuracy while ensuring privacy for search patterns and forward security for files. Additionally, Verifiable Data Integrity (VDI) is achieved through blockchain-inspired Merkle Tree structures. The proposed scheme is rigorously validated using post-quantum cryptography assumptions like Lattice-based Cryptography, ensuring resilience against quantum attacks. Comparative experiments demonstrate that the proposed scheme outperforms three recent methods, achieving superior functionality, precision, and efficiency.