Granica: Scale, Secure, and Optimize Training Data with Cutting-Edge Technology

Rahul Ponnala, Co-Founder and CEO

There was once a prevailing notion that superior AI and ML models depended on amassing extensive datasets, but recent research has revealed that it is not only data quantity that matters but also information quality – and thus information density and efficient cost control.

The path to proficiency lies in the discernment of genuinely relevant data for a specific AI application and the extraction of valuable insights from it.

Enter Granica, an AI research laboratory and systems company dedicated to enhancing AI's quality and performance. Its emphasis on data as a source of information with variable and typically sub-optimal information density, rather than merely a collection of bits, both fuels and cost-controls the era of massive models and highly scalable AI systems.

"Our primary focus is on training data, which we use to amplify AI's effectiveness, safety and efficiency. Our mission is to use a deep knowledge of data, tasks and models to cost-effectively scale the production use of AI within enterprises by a factor of 10,000. It is an ambitious mission, and it drives us to address the entire AI stack, starting with data and extending to algorithms, infrastructure, and beyond,” says Rahul Ponnala, co- founder and CEO of Granica.

Ponnala, with a strong academic background in mathematics and computer science, is dedicated to pioneering innovative products through cutting-edge research. With nearly 20 years of experience in engineering, his early work includes developing algorithms for information retrieval and data security, which remains relevant today.

This knowledge and expertise have been instrumental in the formation of Granica's research lab. Led by Chief Scientist Andrea Montanari, the lab is the driving force behind the company's pioneering and fundamental research in data safety and effectiveness in AI. This specialized division creates techniques and algorithms that optimize AI systems by maximizing the information density within training data sets to in turn maximize model performance and ultimately the insights and outcomes from AI.
The lab addresses the four fundamental pillars of AI in the cloud – infrastructure, algorithms, tooling, and data. It starts by optimizing the use of graphics processing unit (GPU) and CPU infrastructure, recognizing their pivotal role in AI progress. Next, it develops crucial algorithms to boost AI performance, safety and efficiency. The lab also streamlines processes for training and inference.

The core focus of the lab's work is making AI/ML datasets and models both leaner and more robust. This involves increasing information density and signal-to- noise ratio; preserving privacy of sensitive information and PII; and implementing novel data compression techniques. These measures optimize training data storage and processing, elevating model performance and scalability. The research lab is central to shaping a more privacy-conscious future for AI, exemplified by its groundbreaking products Granica Crunch, Granica Screen and Granica Chronicle.

Granica Crunch is a one-of-a-kind training data compression service which improves model performance by storing petabyte-scale training data more efficiently. It harnesses advanced compression algorithms to deeply compresses training files to save up to 80% on storage and access costs, enabling ML teams to re-allocate the savings to acquire and use up to 4X more data to scale and improve model performance.

Granica Screen is a high-accuracy training data privacy service which de-identifies PII and other sensitive information at high scale, unlocking more data for model training and GenAI fine tuning while remaining compliant with data privacy requirements. It champions data security through its privacy-enhancing mechanisms, empowering private machine learning, language models, and AI applications while upholding user data security.

Granica Chronicle is a generative AI-powered training data visibility service which facilitates data-related exploration, access governance, and cost optimization. It enables ML teams to unlock additional budget for reallocation to strategic AI areas such as acquiring and using more training data to improve model performance, investing in people and tooling, and more.

These solutions collectively form its AI training data platform, which over time aims to boost AI capabilities by orders of magnitude through ongoing fundamental research and innovation.
The benefits of these solutions can be seen in a recent instance when the Granica platform played a crucial role in solving a client's challenge of managing a vast repository of data with 100-115 million pre-compressed objects created and stored every day. The client was facing high storage costs. Granica Crunch achieved over 45 percent data reduction on top of the pre-existing zip compression, generating significant annual savings in the order of millions of dollars. These savings enabled the client to control costs by flattening the cost curve associated with training data growth, as well as to invest in more data to improve their AI and model performance.

Our focus centers on the foundational element of AI – data – which we leverage to amplify AI's efficiency and resilience by an astonishing factor of 10,000

To address the client’s data privacy needs, Granica Screen detected and protected sensitive information thus unlocking it for use in advanced AI applications. Granica Screen could protect around 10x the volume of data at the same cost as Google Cloud data loss prevention (DLP) while maintaining high precision and recall. This enabled the client to unlock 10x more data for use in their model training, safely boosting model performance and AI outcomes. The innovative training data platform approach transformed model effectiveness, elevated privacy protection standards, and delivered powerful cost control to the client.

The autonomous, dynamic scaling products it offers are designed to refine data workloads and streamline data protection without disrupting production tasks. The primary offerings, Granica Crunch and Granica Screen, are value-driven, with Crunch pricing tied to cost savings and Screen based on service-level agreement (SLA)-driven guarantees. Granica Chronicle is included in Granica’s base platform fee.

The firm’s mission is to help their customers innovate and achieve efficient growth in fields like drug development, climate impact, and space exploration by significantly improving AI model performance and predictive accuracy in an efficient manner through research-driven, value-focused products. Its unique approach, based on fundamental data principles, sets it apart and positions it for transformative growth in the AI landscape.