Mission
It is well-known that only 13% of machine learning models make it into production. By launching our production-first AutoML platform, we bring that number up to 100%. Located in the heart of San Francisco and Silicon Valley, Koyote Science, LLC, we help companies eschew the need to hire teams of data scientists and engineers to design and deploy their machine learning models, instead allowing them to tap our Bandito™ API directly into production data to get predictions, optimizations and recommendations right away. Our solution breaks down the barrier to entry for both small and large businesses, allowing you to innovate quickly and speed ahead of your competition.
Don’t get bogged down building and managing the following:
Painful alignment of production and offline data for model evaluation
Perfunctory A/B testing design, waiting periods, and evaluations
Lengthy feature engineering and hyperparameter tuning
Dashboard creation and hosting
Our APIs begin optimizing results as soon as they are confident in their predictions, and since our systems extract the maximum utility out of every data point, you don’t have to worry about being in the Big Data business to reap the benefits, although our solutions easily handle Big Data too.
Browse to each model’s hosted dashboard to keep up with how it is performing and gain insights into what works and what doesn’t, then use the administration dashboard to make changes or freeze your model in time for savings and efficiency. With our innovative HyperTune™ technology, we have eliminated the need for feature engineering and hyperparemeter optimization in the vast majority of cases so you are always serving the best possible predictions at all times. Built on AWS Lambda and with the latest engineering and mathematical innovations, our services are cheap, scalable, and secure, to provide you the best results that advanced statistics can buy.
Our technology has already been used in a variety of ground-breaking products, like Smart Slideshow™, which allows Shopify store owners to try out different banner images in real-time to see which ones produce the best conversions. Our software AI guitar educator, Fret Ferret™, automatically personalizes lessons in response to implicit feedback signals from the student to save time and accelerate their learning. Fret Ferret™ is currently being used at the Eastman School of Music in Rochester, New York, to advance its guitar students, and was published at the top conference in computer-human interaction, the ACM CHI Conference on Human Factors in Computing Systems. Niche applications don’t have to go unattended when machine learning solutions are this easy. In addition, we are exploring applications in marketing, sales, education, and consumer products, so see our product vision to learn more about the exciting opportunities ahead!
We are eager to help your organization fly past expectations with Bandito™. With specialized experience in Natural Language Processing and Understanding (NLP/NLU), recommender systems, personalization, signal processing, deep learning, multi-modal learning, hyperparameter tuning, Bayesian optimization, Bayesian inference, and reinforcement learning, our consulting services can unblock your potential. Our leadership has advised the C-suite at top companies like Twitter and Pinterest and designed and contributed production-quality machine learning solutions for Amazon and Alexa. With world-class experience in engineering, mathematics, and data science, we have skillset to help you soar.
Specializations
NLP/NLU
–
We can design and launch a custom NLP or NLU pipeline to give you state-of-the-art performance on your chatbot, search system, or business intelligence
Bayesian Optimizaion
–
We can implement bayesian optimization and inferencing solutions for a variety of problem spaces, such as hyper-parameter tuning, to help you develop quickly and confidently
Reinforcement learning
–
A/B testing is so 21st Century. We can help you upgrade to multi-armed bandits that work together in real-time production, allowing you to launch new products more quickly and with higher customer engagement
Deep Learning
–
We can build our solutions using less resource-intensive algorithms hosted on AWS Lambda, or we can upgrade to high-performance deep-learning algorithms hosted on GPUs using AWS Sagemaker
“We bring a diverse range of experience to solve problems at the intersection of academia and engineering.”
Douglas Mason | Founder
History
Koyote Science was founded in 2017 to build deep learning solutions in the music space. Within six months, we presented a tech demonstration to Amazon Music, and we were immediately asked to join and build out their ML platform. Since then, we have built recommender systems, improved audio classifiers, and dramatically improved music auto-tagging using multi-modal deep learning models. We advised the Amazon Alexa org on how to quickly propagate ML solutions to developers, launched Alexa’s most-accurate and flexible NLU system using Google’s BERT model, and built a multi-armed bandit pipeline using the best available techniques. But most importantly, our solutions are robust, achieving the best balance of depth and breadth by delivering viable products at each stage of execution, such as Alexa’s “help me find a playlist” feature, which was extensively covered in the media.
Recently, Koyote Science has been pushing the boundaries of applied research in Bayesian optimization and inference, and has used the fruits of these efforts to release Fret Ferret™, the world’s most robust guitar trainer that automatically learns from user performance to custom-tailor its lessons. With backgrounds from Harvard University, UC-Berkeley, the Department of Energy national laboratories, and top technology firms — and with a history of releasing impactful real-world applications based on the latest science — we can provide you state-of-the-art solutions that really work.