How Ancestry Decides Which AI to Use to Synthesize Its Data Trove

There’s a fierce battle underway in Silicon Valley, and there’s one piece of technology at the center of it all: AI.

When OpenAI released its AI-powered chatbot, ChatGPT, in 2022, it introduced large language models to the broader public and was embraced by business leaders. It also cast a glaring spotlight on its competitors, some of whom raced to release their own AI-powered chatbots.

The growing number of LLMs poses a tricky question for companies determining which model to use.

Sriram Thiagarajan, Ancestry’s chief technology officer and executive vice president of product and technology, told Business Insider that the company is taking a more-the-merrier approach.

“We are agnostic to LLM models,” Thiagarajan said. “We use multiple AI models. Be it Azure, OpenAI, Meta’s Lama, or offerings under Amazon Bedrock.”

For Ancestry, a Utah-based genealogy company that sells DNA test kits, the AI model’s brand is less important than the end result.

“We built an abstraction layer — what we call an AI gateway — on top that helps us leverage whatever models better fit our case,” he said. “We have built our own agentic framework in a way that helps us provide that unique family story and personalized experience for our consumers.”

Ancestry and AI

Thiagarajan said Ancestry had just begun diving into AI and machine learning when he joined the company in 2017.

At that time, Ancestry was trying to find an efficient way to digitize content, which is a massive undertaking for the company. Ancestry collects numerous forms of records, including birth, death, military, land, immigration, census, and newspapers.

“We’ve collected over 65 billion records across 80-plus countries,” Thiagarajan said. “Just to give a scale, that’s about 10,000 terabytes of data on our platform that we use to provide discoveries to our users.”

Related stories

In the past, processing and identifying troves of records took months.

“About 15 or 20 years ago, when we digitized the 1940 census, it took us about nine months to do it in a manual way at 10 times the cost,” Thiagarajan said.

Then, the company embraced AI.

“We said, ‘Why don’t we apply computer vision AI techniques to automatically digitize content without manual intervention?’ Thiagarajan said. “Fast forward to the 2021 timeframe, we used our own proprietary handwriting recognition computer vision technologies, and we compressed the time to market to under nine days from nine months at a fraction of the cost.”

Still, Sriram said the company relies on humans to fact-check the AI answers “as needed.”

“We’ve built some automated controls and systems that certainly reduce the amount of time we need to spend checking,” he said. “We want to be extra careful in making sure that what we produce using AI is grounded in truth. Grounded in facts.”

These days, Ancestry is going all in on the tech — especially with its employees.

“There are a lot of things we do to drive AI education within the company,” he said. “Be it brown bag lunches, internal forums, or providing an environment where people across departments can experiment.”

Hackathons are also regular events at Ancestry and provide a space for collaboration across teams.

“So encouraging innovation as part of the flow of work, as they call it, as opposed to that residing as a silo,” Thiagarajan said.