I’m a student from San Diego studying data science at UC Berkeley
Aila Health is an early stage precision healthcare start up focused on using machine learning to improve the treatment of chronic conditions, primarily autoimmune disorders. I am currently working on building several ML systems including building knowledge systems to support physicians.
Arricor is a PNW based machine learning consulting company. I am building AI products using advanced NLP for products not yet made public.
My primary goal is to improve the accuracy of medical computer vision systems with advanced data augmentation. I run experiments involving applying state of the art papers to Curemetrix products and data. I also do mare tradition data science work including quantifying the efficacy of our products and comparing it to the efficacy of radiologists. I consolidate this information into reports for FDA approval.
Developed a web application security lab in conjunction with the cyber security and cloud engineering teams. Used lab to programmatically deploy vulnerable web applications to GKE and then attack the applications while monitoring behavior. Later deployed RASP protections over the web applications and measured how much protection each RASP provided. Recommended RASP vendor to senior leadership.
My team at ML@B consulted with climate.ai to prototype a machine learning system which uses geospatial satellite data to predict and analyze crop production in underdeveloped regions. Our prototype has since been pushed into production.
I led a team of 5 undergraduates in a research project through the Data Science Discovery program at UC Berkeley. Our goal was to use DCGANs (deep convolutional generative adversarial networks) to enhance and augment undersized data sets for medical applications. The end goal is to improve accuracy of object detection systems by producing more data from a limited sample size. Our novel approach increased accuracy on the test set by an average of 12.4%.
I built a prototype object detection and localization system to be run in real time for medical applications using YOLOv3 and Mask-RCNN. The project was to help medical professionals find polyps in colonoscopy footage while performing the operation. Achieved frame accuracy of >98% and can theoretically run upto 36 frames per second when optimized.
Assisted UC network engineers with maintenance upkeep and expansion of the campus wide network infrastructure.
Worked in Dr. Shresta's lab during the summer of 2016 studying flaviviruses such as Zika and dengue fever during the Zika outbreak in Brazil. Duties included data collection and management.
Minor: BioEngineering, Certificate in Entrepreneurship & Technology via SCET, Design Certificate from Jacob's institute, member and officer at Machine Learning @ Berkeley
Highest grade in program
Completed with honors in Spring of 2018.
On July 6 2020, the Trump administration announced a policy that said that all international students not enrolled in in-person classes for the coming semester must leave the country. I found this policy obnoxious and so with the help of some friends I made a website. This site allows international and domestic students to connect to ensure that everyone can get the classes they need. The site was up and running within 60 hours of its inception and got over 5000 hits within 12 hours of its launch.
In response to nationwide protests in June 2020, I created Anonymize as an anti-surveillance resource for those who might face retribution for being seen at political events or rallies. I had heard multiple reports of people being harassed or fired from their job because they were in the background of a picture that didn't align with their employers political beliefs and that was later published on a media outlet. My app solves this problem by using facial recognition and images processing to find faces and blur them automatically. This site is still under development.
A common problem in the world of computer vision is building systems that perform as well as humans in conditions that are adversarial to the was machines see. My team and I developed a novel data augmentation and training procedure to improve accuracy of computer vision systems on natural adversarial examples. Report
A known problem in computer vision and image processing is trying to gague distance without any objects of known size in the frame. I used a machine learning based approach to make highly accurate guesses about the distance between a persons eyes based on their demographics and then used prospective geometry to estimate their distance. My approach was accurate within 4% up to 20 ft with my laptop's webcam. github Demo
Natual Language Processing is very well studied in English, but I wanted to see if the same techniques would work in other languages with substantial in syntax and structure. I used traditional approches including Niave bayes, RNN, and clustering to classify and cluster 2000 tweets written in MSA (modern standard Arabic) and Jordanian Arabic. Github
Group project for Data Science in Practice. We build a machine learning pipeline that would scrape data about NBA teams and Vegas betting odds and would apply a conglomoration of algorithms to the data in order to recommend a bet for any given day. The project achived a theoretically small but profitable margin.
I build a real-time object detection and localization to help doctors spot life threatening polyps in colonoscopy footage and other medical procedures. Approach was fine -tuning the YOLOv3 algorithm with hand labeled samples of polyps. System achieved an accuracy of >98% while running at about 5 fps on my laptop. Github