Raki Rahman thinks he uses LLMs more than anybody else he knows, at least among his peers in Toronto.
But the principal software engineer at Microsoft has a complicated relationship with AI tools: he loves their ability to increase productivity but remains a skeptic of their usefulness without tight controls and the right intentions.
“I cannot see a world where an LLM is going to be able to do (...) the thing that I'm doing,” Rahman told The Stack in a recent interview. That “thing” being orchestrating and designing performant, cost-efficient databases to store the ever-growing amounts of telemetry Microsoft tracks.
Rahman works on Microsoft’s SQL Server database team. SQL Server is one of the biggest databases in the world, and Rahman’s team — known as “Telemetry and Intelligence” — gathers real-time data about database performance from the company’s millions of SQL Server customers, whether that's in Azure, competing clouds, or on-prem databases.
“We have real time intelligence, a big data lake, about 300 petabytes plus data,” Rahman said. “My job is to basically make sure that all of our data is getting adjusted, getting processed, and we're able to do machine learning, AI, data modelling, reporting” on the incoming telemetry.
Although Rahman uses LLMs constantly, he would argue against the idea that the technology can easily solve complex business problems, especially in data, which is a large component of his employer's AI product strategy. The vastness of data and the limitations of LLM context windows, in Rahman’s eyes, render the tools useful in specific settings but far from a magic wand.
Rahman shared his approach to LLMs, how he foresees the technology changing data engineering and what he wishes leadership understood about the immense pressure senior ICs are under right now in a wide-ranging interview.
Ownership or bust
