AtishayKasliwal

I finish an MS in Data Science at Stony Brook in May 2026. Most of my work sits in an awkward middle. Half of it is "will this fall over at 3am", half is "is the model even right". I like that middle.

The systems half came from three years at Accolite Digital: an event-driven ETL platform for Fidelity wiring 10+ enterprise systems at 99% uptime, then a Redis and Elasticsearch rewrite that took P99 latency down 40% for 100K+ customers. Zero Sev1s after that deploy, which is the number I care about more than the 40%.

The ML half started at Wake Forest CAIR, where a PyTorch pipeline on GCP hit 90% accuracy and cut clinical processing from twenty minutes to under five across 1,250+ patient cases, with 10TB+ of imaging behind it.

My research at Stony Brook does the same with money instead of medicine. A fault-tolerant pipeline on FastAPI and AWS chews through 200K+ financial data points in real time and drops none of them when the load spikes.

On the side I run Atriveo on my own: FastAPI, React and Postgres, 100+ active users, 2K+ queries a day, and a Chrome extension that has somehow held a 5.0 on the Web Store.

The part nobody puts on a slide is the plumbing. Dockerised CI/CD on Jenkins took our deploy time down 97%, and Prometheus and CloudWatch meant we found the defects before customers did.

What I want next is more of the same, somewhere the systems are big enough to be interesting and someone still cares whether the model is right.