Context
Delta Holding had credit analysis work that was valuable but time-intensive. The opportunity was not to replace analysts with a generic chatbot. It was to make the existing process faster and more usable without removing the need for judgment.
Problem
The core challenge was operational, not cosmetic: analysts still needed useful summaries and recommendations, but the workflow had enough repetition that it was a poor use of expert time. Any LLM-assisted path also had to fit the existing decision process instead of behaving like a disconnected demo.
What I built
I helped shape and deliver an LLM-assisted workflow that produced summaries and limit recommendations across business lines. The point was not only text generation. The point was reducing analyst effort in a way that still fit a real credit workflow.
Production-minded choices
- The workflow was designed around an existing analyst process instead of treating the model as a stand-alone product.
- The focus stayed on usefulness, recommendation quality, and business acceptance rather than novelty.
- The implementation had to earn trust from users, because a speed gain without trust would not hold in practice.
Outcome
The result was a workflow that reduced analysis time by roughly 75%, with analyst quality rated at 4.4/5 and recommendation acceptance above 90%. That is the strongest public signal from this case study, because it ties LLM assistance not only to speed, but to trust and actual workflow adoption.
Why this case matters
This case matters because it shows how I think about internal workflows: identify a high-friction process, design the LLM layer around real user behavior, and keep the system tied to usefulness rather than hype.