As companies are seeing artificial intelligence (AI) continuing to emerge as a focal point, it is important for them to consider how to safeguard sensitive data. It is critical that data is both accurate and secure in order to provide clear results and maximize the potential for AI.
Unfortunately, some large language models (LLMs) are falling short and viewing security as secondary.
“Generative AI can do a lot of amazing things, but only those companies who take security into account at the outset will reap the most benefits.”
— Vaikkunth Mugunthan, Co-Founder, DynamoFL
DynamoFL provides regulation-compliant AI that they claim can help reduce data leakage risks and help companies stay ahead of existing and emerging data regulations while delivering hyper-personalized and effective AI. DynamoFL’s offerings include comprehensive penetration testing for LLMs, rapid deployment of privacy-enhancing AI, and the safe utilization and secure deployment of unstructured datasets. The startup increases performance while reducing the size of data sets and models to help streamline costs.
Vaikkunth Mugunthan, Co-Founder, DynamoFL
Mugunthan and cofounder Christian Lau founded the company in 2021 after completing a Ph.D. from MIT’s School of Electrical Engineering and Computer Science. It’s been a meteoric rise ever since. The company raised $19.3 million in venture capital funding and attracted clients such as Qualcomm, Acuity Knowledge Partners, and Aisin. DynamoFL’s technology claims a 15x reduction in server costs, a 10,000x decrease in data transfer costs, and a reduction in data leakage during LLM fine-tuning from 32.3% to less than 0.01%.
“Our goal is to deploy generative AI LLMs in a trustworthy manner, partnering specifically with enterprises,” Mugunthan explained. “We also offer training and fine-tuning of large models so companies can deploy the best technology possible.”
In the fall of 2023, DynamoFL joined the Comcast NBCUniversal LIFT Labs Accelerator: Enterprise AI, a six-week program focused on helping startups secure pilot programs and strategic partnerships with the organization, where they were able to learn alongside senior executives.
“Our direct interactions with senior executives have been incredibly educational. The opportunity to learn from Comcast and its leaders has been invaluable.”
— Vaikkunth Mugunthan
From MIT to Market: DynamoFL’s Journey
At MIT, Mugunthan and Lau were driven by the shared vision of enhancing data security in AI models. They knew it was critical to move the technology forward, especially in an era where companies are laser-focused on personalization to deliver the right experiences to the right people at the right moments. In fact, companies that excel at personalization drive 40 percent more of their revenue from those experiences than competitors.
Mugunthan and Lau explored federated learning, a method of training AI models without anyone seeing or touching data. After publishing several academic papers on the topic, they transitioned from academia to industry, confronting the real-world challenges of large companies’ data security and privacy regulations. Recognizing the difficulties in data transfer due to privacy regulations, they identified a need for a privacy-centric solution in machine learning.
Mugunthan, who also studied law at Harvard University to deepen his understanding of regulatory compliance, explained: “We believe the key to overcoming data privacy challenges lies in training privacy-preserving machine learning models so they can learn from distributed data sources but make sure that the raw data never gets shipped out and so there’s essentially no possibility of it being exposed.”
The company’s quick success is partly attributed to its development of task-specific models trained on focused datasets, leading to improved performance and significant cost savings from more affordable API integrations and lower server costs.
“We are even getting some of these models to run on mobile phones, that’s how optimized our models are,” said Mugunthan.
What’s Next: DynamoFL’s Vision for AI Expansion
Looking ahead, DynamoFL hopes to double its share of enterprise customers and secure a series B round of venture capital funding to expand its team and enhance its offerings. Big picture, the company’s ambition is to establish itself as the premier solution for private, efficient enterprise AI.
To read more inspiring startup stories and to stay connected to Comcast NBCUniversal LIFT Labs, sign up for our newsletter and follow us on LinkedIn.