Generative AI (GenAI), with all of its potential opportunities, requires a robust data foundation to deliver tangible business value. The efficacy and reliability of GenAI and large language models (LLMs) necessitates intelligent data engineering to ensure that any advanced AI initiative succeeds with high-quality, trusted, and governed data at scale.
John O'Brien, principal advisor and industry analyst at Radiant Advisors, and Abhilash Mula, senior manager of product management at Informatica, joined DBTA’s webinar, Mastering Generative AI With Intelligent Data Engineering, to explore how to best support GenAI and LLM projects with cost-efficient, precise, and effortless data engineering practices.
O'Brien and Mula looked at some of the most pressing questions related to GenAI, its implementation, and mastering the technology with data engineering.
The first question, “How critical is GenAI for the overall success of any enterprise?” is a question that frequently comes up, according to O'Brien. He explained that the generation component of GenAI—whether it’s content, code, or knowledge—is extremely significant, especially as it relates to increasing enterprise productivity in the midst of high-volume data growth. O’Brien predicted that the statistics relating to areas that GenAI improves are not only accurate, but will only increase.
Mula pointed to the fact that according to Gartner, businesses will continue to spend a tremendous amount of money investing in AI technology. With the investment continuing to rise, ensuring that you’re achieving a balance of value will be critical.
Question #2, “What are the major challenges that are derailing GenAI initiatives?” points to the horror stories associated with poor, uncalculated AI adoption.
O’Brien explained that the major challenge he sees revolves around, “‘How do we put in the guardrails? How do we create a secure environment in which we can begin our exploration, our experimentation, and start learning this new technology?’” Answering these questions is entirely tied to defining the use case for utilizing GenAI, he added. Simply jumping on the AI bandwagon and hoping for the best—despite its radical costs and potential downfalls—will fail to generate any sustainable, provable value.
Mula echoed O’Brien’s points, explaining that “the success of any enterprise, especially with AI projects, depends on how effectively we are embedding these AI/LLM models into the core business objectives. If there is no real use case, these projects are doomed to fail.” He further added that with GenAI come concerns for privacy, security, and relevant skill sets to drive its success.
Moving onto the third question, “What role does data engineering play in the success of GenAI?” O’Brien highlighted that data engineering teams and architects will be largely responsible for the environments that GenAI is supported by. Ensuring transparency, accountability, the protection of intellectual privacy, and more, will be the lofty load on the shoulders of data engineers.
However, data engineers are the key benefactor of GenAI, according to O’Brien. Chatbots acting as an assistant to the engineer, pointing out repetitive functions or increasing the quality of the code, are a unique selling point for data engineering teams.
Mula stated succinctly: “AI needs data, and in order to give clean data to the AI, they need to engineer the data. Data engineering is very, very important because AI relies on high-quality, well-organized data to function effectively.”
Garbage in is garbage out, Mula explained further, emphasizing that data engineering is an incredibly critical component of ensuring that any AI project is turning out accurate, useful information.
Ultimately, the key to mastering GenAI with data engineering boils down to a few crucial points:
- Explore AI safely by defining upfront policies and environments.
- Identify use cases where GenAI has the potential to make measurable improvements.
- Ensure that GenAI solutions are explainable, traceable, and able to be optimized.
For the full, in-depth roundtable discussion, you can view an archived version of the webinar here.