What if business teams could get the data they needed without waiting for an analyst to run complex queries? That was the vision Ashutosh Kumar and his team in Data Engineering brought to life at MPL’s very first Demo Days — leveraging the power of Large Language Models (LLMs) to simplify access to data.
Ashutosh, who has been with MPL since 2020, was one of the earliest members of the Data Engineering team. Over the last five years, he’s seen the function grow from the ground up. Today, his team supports analytics, data science, product, and more—any function that relies on timely, accurate data.
At Demo Days, Ashutosh and his team Om Prakash , Devang Patel and Shaik Khaja Nawaz, showcased a prototype that could change how business teams interact with data: a natural language-to-SQL engine powered by LLMs. “The idea was simple,” he explains. “Business folks often need metrics to make decisions, but they have to wait for analysts to write SQL queries. This slows things down. What if they could just ask their question in plain English, and the system generated the SQL automatically?”
Turning a Real Problem into a Demo Days Project
Interestingly, the idea came just 15 days before Demo Days. “Kabir Rustogi (who heads Data at MPL) brought up the issue and suggested we work on a solution,” says Ashutosh. It was a real problem the team faced—making it the perfect fit for Demo Days, which encourages employees to build and present innovative ideas that solve real challenges.
The prototype was built in just under a month. While it wasn’t production-ready, it was a working model for one product pod—enough to demonstrate the potential. “We knew that if it worked for one pod, we could scale it across others,” he adds.
The Real Challenge: Data Modeling
As they built the engine, a deeper issue emerged: the data itself. “Some of our data is scattered. There’s no standard table structure, and no central definition for metrics. The same number can exist in multiple tables,” Ashutosh points out. This made it difficult for the LLM to pick the right source, often leading to inaccurate results.
The team quickly realized that building a robust data model was a prerequisite. “LLMs can’t work on tribal knowledge,” Ashutosh says. “We needed to document everything—tables, metrics, definitions—and start creating a single source of truth.”
This led to a pivot. The focus shifted temporarily to foundational work around data modeling and cataloging—essential groundwork to enable LLM-based tools to function accurately at scale.
Mentorship and Rapid Learning
One of the highlights of Demo Days for Ashutosh was the weekly mentorship sessions. “We were building something for the first time. We didn’t know what a good model looked like. The feedback we got every Friday helped us refine our approach.”
One crucial piece of advice they received was to implement a Retrieval-Augmented Generation (RAG) system. Their initial approach relied on prompt engineering, but as data grew, the prompts became unwieldy. RAG helped streamline the process and improve model performance significantly, just in time for Demo Days.
Budding innovators
Ashutosh believes Demo Days will inspire more people across MPL to come forward with ideas—especially those outside tech roles. “The event pushes you to take an idea seriously. Without Demo Days, we might not have prioritized this project.”
His suggestion? Create a centralized list of real problems faced across MPL. “Many people don’t know where to start. If we had a shared problem bank, folks could pick something relevant to their domain and start exploring solutions.”
Having spent his career in data—working with credit bureaus, building ranking models for document search, and now exploring AI—Ashutosh knows the value of solving real-world problems through technology.
His biggest takeaway from Demo Days? “LLMs have massive potential. If used right, they can drastically reduce the time it takes to make decisions. AI is here to stay—we just need to learn how to use it smartly.”
Demo Days is MPL’s internal platform for innovation—where employees across the company pitch, build, and showcase solutions that solve real business challenges. Stay tuned for more stories from our first cohort.