The company had been working on an integrated prototype called VFNLU, which was no longer an academic exercise but a feasible product. After refining the model, they refactored their code to fit existing systems and added multilingual support. However, they soon realized that their ML platform's latency was too high, leading to a 3-month code refactor to redesign how their NLU dealt with requests. The company then started beta testing the VFNLU in April 2023, rolling it out to free users and finding bugs and deficiencies along the way. After 18 months of work, they released the VFNLU, which outperformed popular NLUs on benchmarks, supporting 30+ languages and having low training time. The company aimed to allow people to experiment with both NLUs and LLMs in the conversational AI world.