Research Assistant, Professor Richard Zemel - Aug 2023 – May 2024• Implemented 5 NLP pipelines to demonstrate the effectiveness of Prompt Risk Control, a framework for making provable guarantees about the safety of large language model outputs, resulting in accepted ICLR 2024 and NeurIPS 2023 submissions• Scaled up and orchestrated our large language model (e.g., 40 billion parameters) pipelines on clusters of graphical processing units (GPU) enabling us to run our experiments in time to meet the conference deadline• Defined clear notation that we used throughout the paper to communicate our framework and co-wrote the paper, resulting in clear scientific writing and a submission that received positive peer reviewsResearch Assistant, Professor Kathleen McKeown - Sep 2022 – Oct 2023• Developed a novel approach for dialogue analysis using circumplex theory for the DARPA Computational Cultural Understanding (CCU) program resulting in a first-author acceptance to LREC-COLING 2024• Collected a novel dialogue dataset annotated with features from circumplex theory (e.g., Gregarious-Extraverted, Aloof-Introverted, etc.) using GPT-4 and trained state-of-the-art dialogue outcome classification models on 2 dialogue datasets• Delivered 2 dialogue analysis systems to DARPA and represented Columbia on weekly calls with all program stakeholders (DARPA, NIST, SRI, Monash, LDC, NYU, PARC, and LCC) resulting in Columbia’s funding getting extended in the second phase of the program

Automated Knowledge Graph Construction• Implemented and compared 7 methods for extracting entities and relations from arbitrary text• Developed procedures to link identified entities to Wikidata entries• Authored and submitted entity extraction paper to the AKBC conference [uskb-workshop.github.io/abstracts.html]Information Retrieval and Classification Model for a Bank• Managed a team of 6 at PwC and collaborated with CMU to build machine learning models to classify text, extract key phrases, and compare semantic similarity of text• Defined data annotation process, end-to-end ML pipelines, testing strategy, CI/CD pipelines, git flow process, and deployed models to cloud platforms to serve predictions• Reduced time spent by compliance experts by up to 50% [bit.ly/32f2lx3]• Listed as lead inventor on 2 patent filingsEnterprise Search• Reduced number of web clicks required for users to find information in PwC’s firm search portal• Deployed 2 machine learning models to Kubernetes (QA model, IR model) and improved search experience for 50K+ users with assistance from 2 junior data scientists• Presented work at GCP Next ’18 [youtu.be/63EANkPzuJY]Select Projects• Led: EEG Based Brain-Computer Interface [youtu.be/f81T0KcprpM], Structural Causal Model Development with CMU Students, Automated Customer Service Case Resolution for a Major Food Chain, Deep Learning on Edge Devices [youtu.be/iqVKjye-J68], Accounts Receivables Forecasting Model for a Technology Firm, Information Extraction from Excel & PDF Documents for Insurance Industry, Markov Chain Predictions for an Auto-Manufacturer, Dynamic Pricing for an Online Ticket Sales Platform, Sales Calls Analysis for an Asset Management Firm, Big Data Analytics for a Soft-Drinks Manufacturer, Classification Model for PwC Auditors, Classification Model & Big Data Analytics for a Brokerage Firm