Blockchain-enabled smart contracts execute without third-party involvement when specific conditions are met. This system significantly reduces the use of paper-based processes while managing substantial funds. As such, it has become the target of hackers. It has suffered several attacks, particularly the DAO (Decentralized Autonomous Organization) attack and the Parity multi-sig wallet hack, which had severe consequences. This has triggered the need for better security measures as the use of this technology has increased significantly in recent years. Several tools have been developed to tackle various problems while they still lack better achievements. As deep learning achieved better results in vulnerability detection domain, we proposed a deep-learning approach using bidirectional long-short term memory (BLSTM) to classify contract either vulnerable (detection of suicidal, prodigal and greedy contracts) or non-vulnerable. The aim was to propose a detection technique that might achieve better results so as to encourage the use of deep learning techniques. The experiment was conducted on a HPC (High-Performance Computing) cluster where each server has a processing power of 2.2 GHz and a RAM of 128G at least. We were able to achieve an accuracy and F1 score of 94%. This works also set a new benchmark to better smart contract detection while promoting future researches axes.