Bachelor's degree

Master's degree

Migrated data to updated website for annual conference
Formatted content to match existing design and layout in HTML
Produced video in After Effects to demonstrate the capabilities of a new software to clients

Software Engineering Intern

Developed communications protocol for Mainframe desktop in Javascript
Expanded App to App Communication within Zowe, part of the Open Mainframe Project. 
Created Zowe applications using Angular and React.

Deployed enterprise grade, non-cloud native DAST scanner to AWS EC2 within Capital One VPC for scheduled application scanning, enabling 2x frequency of scans, saving $16,000 / year (Terraform, Docker, AWS ELB, AWS RT53)

Built web app for employees to schedule DAST scans and view and filter results of scans (Angular.js, S3)

Developed and published Swift iOS app to aggregates positive experiences to improve mood
Implemented user login with RESTful APIs and encrypted database server

Octtone is a mobile application which seeks to master audio in a spatial format commonly referred to as 8D. Spatial audio creates an effect that is often described as music circling the head and is considered by many to be a more fully immersive listening experience. This is done through an algorithm which compresses audio into nodes with space and time components and moving these nodes algorithmically to create the circling effect. The nodes are moved through various mathematical functions with respect of time. The application uses a library of unmastered songs and downloads them on to a user’s device. The application then formats the audio so it can be manipulated by the algorithm into 8D. During playback, the effect is generated on device in real-time, allowing any song to be converted to this style.

Conducted creative design, user experience design, and full stack development of complete meditation
application, Bluezen. Produced custom guided meditations and marketing materials wi

Co-Founder

Weekend side project - AI Infographic Generator

60000+ users

Flask, React, AWS, Prompt Engineering, SEO Optimization

Founder

Developed multilabel content moderation model achieving equivalent accuracy and 4x faster E2E latency, saving $2,000,000 / year (DeBERTa V3, Azure ML)

Created multilabel content moderation dataset with over 500,000 samples.

Applied state-of-the-art machine learning techniques, conducting highly scalable & distributed experiments to generate pseudo labels and denoise labels (confidence learning, pytorch, distributed cloud computing)  

Model scheduled for deployment to Azure OpenAI, preventing intentional misuse and unintended harms for GPT-3 as a service.

Machine Learning Intern

Lead Model Developer for Microsoft Prompt Shields

Supporting 20B+ classifications per year
Saved $6,000,000 / year by unifying user + cross domain prompt injection into a single model

Developed Microsoft Prompt Shields from Ideation phase thru Private Preview, Public Preview and General Availability. Used across multiple products, including M365 Copilot, MS Defender, Copilot for Security, and many more. 

Prompt Shield is available to public via Azure AI Content Safety service (AACS) and Azure OpenAI (AOAI)

In my role, I was the sole ML research scientist responsible for e2e development of two classifiers. This included ideating improvements, creating new datasets, writing distributed computing code, running scalable and deterministic experiments, selecting release candidate models, and deploying to production.

Machine Learning Engineer

Evaluated the potential of unsupervised and supervised learning to reduce false positives in an enterprise static code analysis tool, Klocwork (Python, Keras, scikit-learn, numpy, pandas)

Performed data collection, data preprocessing, model development, and model evaluation in Keras

Utilized Shapley values to visualize feature importance and understanding model criterion in blackbox algorithms

Avi Arora

Building Oddpool.com | Data Layer for Prediction Markets