Model Architecture Research — Training

Multimodal code-generation for web data extraction at scaleBacked by Paul Graham, General Catalyst, AI Grant (Nat Friedman & Daniel Gross), SV Angel, Y Combinator, and Rebel Fund (founders of Reddit, Instacart, & Cruise)

Built and shipped a research tool for content marketing that helps SEO writers know what to write about• Conducted applied NLP research for document clustering, information gain, and intent filtering• Built backend & infra to serve MVP to 100+ monthly active users• Worked with B2B customers paying >$25k to identify pain points & iterate on key features• Interviewed & rejected for YC S23, decided to work on something bigger, bought out by co-founder

• SCS Senior Thesis: "Semantic Composition in Visually Grounded Language Models"• Exploring compositional ability in multimodal models with Paul Liang and Prof. LP Morency in Multicomp Lab• Discovered novel state-of-the-art self-supervised learning approach for vision-language compositionality

• Used semantic graphs to probe neural compositional ability in LM's with Uri Alon and Prof. Graham Neubig in NeuLab

• Worked on Azure Machine Learning team within AI Platform group• Fine-tuned & distilled language models for named entity recognition• Developed serverless ML assistance tool for natural language data labeling

• Developed discourse graph model for topic-based abstractive summarization• Researched & deployed neuro-symbolic solutions to meet customer requirements

• Optimizing grammar & reinforcement learning library under Prof. Adrian Brasoveanu• Developing RL algorithms for cognitively faithful syntax parsing

• Training sequence classification models under Prof. Marilyn Walker and Kevin Bowden• Developing conversational NLP flows for Amazon Alexa Research Prize 2021

• Developing deep learning models for Brain-Computer Interfacing and NLP• Due to research confidentiality constraints, further details TBA• Intheon creates neurotech tools for research labs around the globe• Using TensorFlow, NumPy, Python, and AWS

•Implemented TensorFlow.js computer vision models in the browser•Built React web app to calculate force exertion of humans on video•Product to be deployed in gyms to replace >$20k in equipment

• Fine-tuned PyTorch NLP models for question answering (F1 >0.9)• Implemented neural and classical information retrieval approaches• Productionized with Flask REST API for core company product

• Teaching multiple students aged 10+ • Courses include introductory programming, web technologies, and machine learning• Technologies taught include Python, JavaScript, HTML/CSS, Flask, React, SQLite, and PyTorch

• Architected high-dimensional recurrent document regression model• Collected and cleaned >7 million textual data points from Internet• Examined relevant academic literature and wrote preprint• Research conducted under Prof. Narges Norouzi at UCSC

• Developed Pulse Navigation's mobile app and dual wristband • Used Dart (Flutter), C++, Google Cloud Services, Bluetooth LE, microcontrollers, RF transceivers, and 3D printing• Assisted in market research, product testing, and video production in preparation for Kickstarter launch

My venture Brilliantile sold outdoor tiles that remove snow based on weather data. I developed our IoT software as well as some of our hardware.

• Built React app that dynamically loads thousands of database objects• Coordinated migration from Google Realtime Database to Cloud Firestore• Integrated Google Firebase authentication