The market is one of the most fast moving and elusive systems on the planet. At peak hours, ticks come in at 300,000/sec. As a company which promises financial data, Bloomberg needs a way of establishing metrics on data latency and correctness. I am on the Market Data Latency and Correctness team, where we are building engines that calculate market latency end to end and estbalish personalized, measurable correctness algorithms. The insights generated from our team help to guide initiatives, cater SLOs, and address issues in the Bloomberg data pipeline, which forms a large portion of annual revenue.

Every body is different. The more insights that doctors have on a patient, the more they can cater surgical procedures to the patient's speciic needs. At Kelahealth, I worked to create a surgeon dashboard, fed by a predictive analytics model, which displayed patient risks and procedure recommendations. The dashboard was demo'ed at 11+ hospitals. Throughout, I worked to integrate customer feedback and ever-changing backend data schemas/structure to provide an efficient backend and seamless and intuitive frontend experience reqired in such a high-stakes environmnet.

Advertisements are ever-changing and must connect with every age, gender, location, and interest. At Publicis Sapient, I worked to create an advertisement dashboard powered by generative AI, which would assist McDonalds marketing in the ad creation process. My responsibility was the backend, which included creation of the customer segments based on their McDonalds profiles (gender, location, point value, buy frequency, etc.) and prompting of generative AI models to create advertisements based on the conjunction of customer segment and local events (i.e. concerts, sports games, holidays, etc.). I additionally dseigned the API which fed the frontend dashboard. These results were presented at the end of the summer to relevant Product Managers and staff.

To make a map of the brain and all of its neuronal connections (a connectome), researchers have begun using flood-filling computer vision models on electron microscopy images of sliced cortical segments which produce continuous segments through space. However, these algorithms are prone to errors, so I worked to correct errors on the output of the algorithms through proofreading and use the Neuvue platform. Proofreading is tedious task, which requires lots of human input, and so my team sought to game-ify the process by wrapping the Neuvue platform in a game. Users play as an astronaut and fly through the "universe" to make paths, when in reality they are flying through 3D representations of electron microscopy images, and the paths they are making are continuous segments of dendrites and axons. Results pertaining to the Neuvue platform were published and presented in a symposium.

Researchers know that when you learn a task, neurons exhibit adaptive behavior, modulations in their signals, throughout the learning process. This research sought to examine how neuron clusters are recruited in an auditory learning task in mice. We constructed computational models to understand relationships among neurons (i.e. are they a part of the same subspace). I worked in Python and Matlab to create an analysis pipeline for dual-recorded electrophysiology data. This involved correcting for rig anomalies, consolidating waveforms, and visualizing and interpreting their shapes and locations to determine their role within the learning task.

Single cell RNA sequencing pipelines involve the identification of cell types (here in tumors), based on expression profiles obtained from the isolation and sequencing of RNA. Data analysis pipelines create reports that are helpful for researchers, but the normalization, PCA reduction, clustering, and visualization can be parameterized in ways that skew results. In this project, I designed data analysis pipelines and modified normalization and visualization hyperparameters to examine downstream effects in the ultimate cell type results. I discovered that results are far from definitive, and depending on user input, potentially unique cell clusters could be consolidated. I completed this research beneath Dr. Uma Chandran and presented these results at a symposium at the Hillman Cancer Center.