itssujeeth Distinguished Architect / Technology Director · 2024 Delivered
LLM-Assisted Requirements Analysis
An LLM-assisted approach that extracted, summarized, and correlated requirements from a large volume of recorded discovery sessions inside an enterprise modernization program, on a schedule manual analysis could not meet.
A legacy i-Series/RPG invoice modernization program moving to GCP microservices depended on a business requirements document that two analysts had 45 days to produce from 29 one-hour recorded sessions. Manual review of the material would not finish in time, and this was the organization's first use of generative AI on its own business data, so cost and data handling had to be addressed before anything ran.
I proposed and implemented the approach: extraction, summarization, and correlation of requirements from the discovery material, run inside the customer's own cloud environment. I addressed the cost and data-handling questions a first use of generative AI raises, and designed the workflow so that business analysts verified every output against source evidence and remained responsible for the final requirements. I also supported solution design and design reviews for the wider program.
Delivered: the business requirements document was completed within the 45-day window from 29 one-hour recorded sessions. Estimated by me: about four weeks of analysis and correlation work saved for each of two analysts. Follow-on: the customer began encouraging LLM use elsewhere, and by Q3 2026 had four agentic AI solutions in progress across several portfolios.
This is a small example of the judgment I think matters most in enterprise AI: spotting where a model can genuinely help, introducing it without creating a new risk, and keeping people accountable for the outcome. The model drafted and correlated. The analysts decided. That split is the same principle I apply to agents that touch money or operations, applied here to a document.
Running inside the customer’s environment was not optional. A first encounter with generative AI on real business data sets the organization’s expectations for every project after it, so cost transparency and data handling had to be right the first time.
Client and employer details are confidential.