Jul 2024 – Present · 1 yr 11 mos · Remote

• Spearheaded the causal analysis of U.S. sports betting’s impact on gambling addiction via synthetic control and time
• series forecasting; quantifying a 23% nationwide increase in addiction.
• Engineered ’Waldo’, an open-source NLP tool by fine-tuning the RoBERTa LLM on 360,000+ Reddit posts to detect
• Adverse Events with 99.7% accuracy, surpassing GPT-3.5 Turbo by 57% in F1-score.

Dec 2022 – Aug 2023 · 8 mos · Remote

• Created a new dataset by utilizing advanced image augmentation techniques to generate high-quality synthetic data, thereby expanding the dataset size to 100,000 images (30% increase), improving robustness of the model.
• Tailored a Vision Transformer architecture to address edge-case scenarios in traffic sign detection, beating the state-of-the-art F1 score by 9%

Aug 2022 – Apr 2023 · 8 mos · Hybrid

• Led A/B testing of a hybrid music recommendation system; proposed model (treatment) resulted in a 33% higher
• click-through rate (CTR), 20% longer session duration, and a 25% increase in app adoption among target users.
• Deployed the recommendation system into the live application using AWS Sagemaker with continuous updates for
• improved performance, collaborating cross-functionally to ensure seamless integration and scalability

Jan 2022 – Jun 2022 · 5 mos · Remote

• Spearheaded a team of three in building a YOLOv5-based real-time object detection model for an automated food
• delivery conveyor belt, achieving 98% detection precision and increasing customer footfall by 30%.
• Collaborated with the hardware engineering team to synchronize the belt’s speed with the model’s processing time
• using Raspberry Pi, reducing false negatives by 15%; later reused for inventory tracking, reducing food waste by 20%.