In a significant development for the field of AI validation, Bobidi and MLtwist have published a case study demonstrating 50% cost savings, 90% faster deployment, and improvements in data quality through automated data pipelines for AI. Founded by Google and Meta veterans, Bobidi has been leading the way in AI validation, providing cutting-edge tools and expertise to help companies ensure the accuracy and efficiency of their audio AI algorithms.
Bobidi’s challenge: Validating 5,000 audio files, totaling 10gb of data, with 600,000 existing audio labels in under a month. They also needed to generate an additional 50,000 new audio labels. To accomplish this, they needed to perform 100,000 file transformations in a 9-stage process and build and maintain a project-specific data processing pipeline.
By leveraging MLtwist’s technology, Bobidi was able to implement an out-of-the-box pipeline, thereby dramatically reducing the time required for validation. MLtwist’s scalable AI data pipelines enabled Bobidi to push data to the AI tools they needed, resulting in a 90% faster deployment for Bobidi and 50% savings of their allocated budget, equivalent to $25K/month or $300K/year.
Furthermore, the processed data exceeded quality requirements, with an accuracy rate of 98%, greater than the 95% target thanks to MLtwist’s proprietary quality control process.
“Thanks to MLtwist’s AI data pipelines, we were able to reinvest 50% of our data science spend to increase model performance, while the labeling quality beats any existing open source or commercial solutions we have tried. The accuracy rate is particularly impressive given the large volume of data being processed,” said Dr. Soohyun Bae, CTO of Bobidi.
About Bobidi
Bobidi offers an AI model test platform for AI companies to safely validate models before deploying them. Bobidi leverages a global community of people to test models and find biases, which makes the entire process 10x more efficient.
About MLtwist
MLtwist turns raw images, video, audio, text, sensor, and 3D security data into human-verified training sets — on its team, the customer’s team, or both.
Thanks to MLtwist’s AI data pipelines, we were able to reinvest 50% of our data science spend to increase model performance, while the labeling quality beats any existing open source or commercial solutions we have tried.” Dr. Soohyun Bae – CTO, Bobidi