Credit to Thomas Booska (@vulpesjr) and Julian Rice (@pooska)for the photos

• Responsible for the machine learning/ NLP parts of the eligibility system.

• Meet in daily developer meetings with the rest of the team to give updates

• Wrote code that predicts if a item is eligible or not using Machine learning and NLP approaches.

• Deploy ML and NLP models to work with the Supabase backend using Google Cloud Run, Flask, and Docker.

• Proficient at using retool with sql queries to display visualizations and statistics about the dataset.

• Design and built a data pipeline using Supabase edge functions, database triggers as well as a Google cloud run model.

• Worked as a member of the data acquisition team and helped build the Atlas DAQ system that transferred data between various rocket engine sensors and components (thermocouples,load cells, strain gages, and pressure transducers).

• Wrote classes in python that interacted with these sensors (mainly thermocouples) through a Labjack T7 to fetch their measured value so that it can be gathered by the overall DAQ system code that collected readings from all the various systems connected to the DAQ system.

• Performed sample rate testing for the sensors and added resolution/range settings in the main github code for these sensors.

• Wrote visualization functionality using Pyqtgraph package in Python to visualization the sensor data fetched from the DAQ.

Supervisor: Professor. Reihaneh Rabbany

Intern research assistant for the complex data lab.

• Engineered prompts for gpt3.5 that was used to generate a human trafficking NER dataset (HTGen)

• Exploited LLMs to generated synthetic annotations for a cost effective way to get labels on niche domains as well as domains where the data is highly sensitive and thus not perfered to be labeled by human annotators.

• Explored rule base gazetteers and knowledge bases of human names.

• Continual learning to expand existing knowledge bases

• Surveyed the ethics (bias and toxic lanugage) in GPT generated synthetic human trafficking domain data.

Papers:

1. SWEET - Weakly Supervised Person Name Extraction for Fighting Human Trafficking

Conference: EMNLP Findings 23

DOI:10.18653/v1/2023.findings-emnlp.219

2. T-NET: Weakly Supervised Graph Learning for Combatting Human Trafficking

Conferences: AAAI-24: AI for Social Impact (AISI)

DOI: https://doi.org/10.1609/aaai.v38i20.30233

3. SWEET - Poster Presented

ACL2023 Student Research Workshop

Link: https://virtual2023.aclweb.org/paper_S122.html

Volunteer positions:

Student Assembly Note Taker

Montreal AI Symposium Reviewer and volunteer