A 30-Day Analyst Process Cut to Five Seconds

// CASE RECORD

The problem

A venture firm curated millions of deals through a process that took its analysts thirty days by hand, every cycle. Behind it, twenty-plus data-ingestion jobs ran out of a million-line proof-of-concept Java monolith that needed a day and hundreds of virtual machines to finish.

What I did

As lead software engineer I designed and deployed distributed streaming microservices for the deal pipeline and directed the pod’s information architecture with the front-end engineers shaping the API. Working directly with the analysts, I built services that turned the thirty-day manual process into a five-second AI-driven task followed by three days of human review — the people stayed in the loop where their judgment mattered and left where it did not. I then broke the ingestion monolith into twenty-plus streamlined processes.

The result

Thirty days to five seconds plus three days of review. Ingestion that took a day now finished in one to four hours. Hundreds of virtual machines became fewer than ten. The method is the one I still use for automation work: start from the workflow the business already measures, automate the repetitive core, keep review where it earns its keep, and report the before and after in the business’s own units.