Work Research infrastructure that has to actually run. Stony Brook University, November 2024 to May 2026.

An async, event-driven backend that ingests over 10,000 FOMC intervals, production AI/LLM services on top of it, and the dashboards that make the output usable for a research team, not just for a notebook.

Where: Stony Brook University · New York
Dates: November 2024 – May 2026

10K+FOMC intervals ingested
62%directional accuracy
80%cut in inference timeouts

Stony Brook's campus quad in autumn, fountain and academic buildings in view

01 · An async backend for a research pace

The platform ingests more than 10,000 FOMC intervals through Django REST and asynchronous processing on NoSQL storage, built to enable macro prediction models that reach 62% directional accuracy.

02 · Production AI, not a demo

LLM service components run in production with connection pooling and circuit breakers, cutting inference timeouts by 80% and stabilizing 10,000+ daily embeddings for live NLP signal generation.

03 · Interfaces for people, not just for me

React and D3.js front-end features give more than 20 people across 3 departments real-time data visualization with sub-2-second performance, so the research is legible to people who didn't build the pipeline.

04 · Iteration speed as an engineering problem

DevOps changes, Docker, CI/CD, CloudWatch, took one experiment loop from 6 hours to 90 seconds, at 99% uptime. That change alone is what made the rest of the work practical to iterate on.