In 2015 Indico shipped its first large-scale language model into production. Since our founding, we have stayed on the bleeding edge of leveraging the transformative effects of modern neural networks, singularly focused on applying AI to help our customers solve their unstructured data challenges. Modern AI techniques have redefined the state of the possible for unstructured data, but massive compute loads, non-traditional data pipelines, and a firehose of research makes the adoption of these technologies cumbersome, error-prone, and ultimately impractical for most enterprises.
Recent studies show that fewer than 11% of companies have successfully captured returns from their AI investments. By comparison, our customers experience a 97% success rate. A combination of technology, applications and best practices enables our customers to capture rapid time-to-value, manage their deployments at scale, and drive bottom-line impact.
Composite AI is about combining modern AI approaches including neural networks with a range of other AI approaches like rule-based reasoning, graph analysis, transfer learning, and machine teaching. The goal is to enable AI solutions that require less data and energy to learn and which embody more “common sense” approaches to model creation. Composite AI recognizes that no single AI technique is a silver bullet. Indico’s approach leverages a powerful set of techniques.
Even the most sophisticated traditional AI systems are constrained to a single data mode, such as text or image. Indico can reason across data modes, incorporating visual information alongside semantic information to make decisions. This results in significantly higher accuracy for multimodal use cases like document understanding.
Indico is the industry leader in the adoption and application of Transfer Learning. Transfer Learning allows customers, for the very first time, to build high-quality custom machine learning models with as few as 200 labeled examples. Transfer Learning allows customers to effectively “amplify” their small labeled datasets to perform as if they have labeled tens of thousands of samples.
One of the biggest challenges in AI is training data. The ability to accurately capture real world data is critical to the creation of high performance machine learning models. Machine teaching is, at its core, a human-centric approach to AI. Rather than asking “how do we learn?”, we ask “how do we teach?”
The world is constantly changing and even the best models will be out of date in months. Indico takes a Staggered Loop approach to Continual Learning. As data passes through your production models, Indico captures user corrections and feeds them back into staged models for evaluation and then to production.
With nearly a hundred custom tasks, tens of billions of words, and terabytes of images, Indico has the most extensive benchmarking suite of any company in the industry. We test the river of bleeding-edge research from top institutions in the world to ensure that our customers always have access to the state-of-the-art with our open-source Enso project.
With a built-in 80/20 Train/Test split, all Indico models are trained using data science best practices, including preventing common pitfalls such as overfitting. Our Explain module includes a set of best-practice metrics such as F1, Precision, Recall, ROC/AUC and more. Our model training dashboard includes troubleshooting capabilities including class imbalance to help diagnose and improve models.
When 90% accuracy just isn’t enough. Indico’s human-in-the-loop interface lets subject matter experts take the reins. Tailor our Review Queues to fit into your use cases with a composable interface that can integrate with your favorite enterprise platforms.