Requests
0
Keep in touch with meI'm using Intch to connect with new people. Use this link to open chat with me via Intch app
Work Background
Booz Allen HamiltonComputer Vision / Machine Learning Engineer
Aug. 2019アメリカ合衆国 メリーランド ローレル- Replaced existing model for thermal data with an improved detection model on thermal data, raising mean average precision score from 25% to 75% - Wrote scripts to generate shared object files from packages for Python to C++ to improve runtime performance by 10% - Developed and deployed machine learning models using containerized workflows with Docker-based infrastructure - Implemented a simulator for testing embedded software on a desktop environment, significantly improving workflow of team members - Trained a vehicle classifier in Pytorch hosted on device with constrained memory to recognize make and models on hundreds of classes (160) on imbalanced dataset (improved team's model from 71% to 95%) - Implemented a CNN multi-object tracker based on Deep Q-Learning from scratch in Pytorch. Iterated on the model through experiments to improve performance - Used image-to-image translation GAN to translate images from thermal to visible domain - Made an auto-annotation data collection system along with entire pre-processing pipeline for generative AI pipeline - Developed robust backend systems and APIs to handle model inference, data processing, and result visualization from deployed ML models - Optimized and miniaturize ML models to be deployed on edge devices with limited memory for autonomous systems / robots considering speed vs accuracy trade off - Developed multi-algorithm orchestration for a sensor fusion solution that unified vision, audio, and radar data processing in autonomous systems, significantly reducing amount of false positives / negative detections by 20%
Prudential FinancialSoftware Engineer
Jun. 2018 - Aug. 2018アメリカ合衆国 New Jersey Roseland- Implemented canny edge detection and morphological transform in order to search for whitespace or less detailed spaces with an image to write informative headlines or text on to it. Removes the need for content writers to go through endless of photos and manually edit and add text, saving content writers a lot of time. - Created an article summarizer that searches for top ranking sentences based on word frequency and presents each articles as serveral main points. Enables employees to catch up on news on Prudential in seconds as opposed to sitting for minutes reading each article. - Made a web scraper along with headless browser scripts to automate the process of generating content and publication for digital signage. Goes through multiple websites owned by Prudential and scrapes images, titles, and article content to create content to be used in digital signage. - Used Jenkins to automate a Flyway database migration pipeline that runs SQL scripts, monitor the process throughout the pipeline, and notifies people if any build problems or errors comes up. Enables developpers to quickly catch errors as they happen as opposed to the next time they run the scripts and check - Designed the UI and server for a certificate verification form using Angular and Node.js
Freelance WorkFreelancing Software Developer
Sep. 2016- Leveraged deep learning algorithms to create efficient soccer ball tracker for DribbleUp, making its way into a production environment.
- Developed deep learning speech recognition model that uses spectrogram data to determine if patient has stroke or no stroke. - Designed pure computer vision solution to create real-time clay pigeon tracker & hit/miss recognizer, currently being in active use.
Texas Tech UniversityAI Research Assistant
Jun. 2016 - Aug. 2016Lubbock, Texas Area- Participating in the Texas Tech University, Declarative Programming Research Program for artificial intelligence. - Designed a multi-agent intelligence system (swarm intelligence) where a team of ally agents work together to surround and capture a fleeing enemy agent. Ally agents searches for and pursues interest points around the enemy agent while minimizing communication between each other in order to catch the enemy agent, which uses obstacle avoidance based on artificial potential fields. - Abstract accepted at the NCUR 2017 conference - Also implemented planning algorithms using declarative programming (SPARC) in problems such as blocks world and vaccine recommendations.
Rutgers UniversityComputer Vision and Machine Learning Research Assistant
Sep. 2015 - Jun. 2018- Participated in the Aresty Research Assistant Program for research in computer vision, machine learning, and robotics. - Got autonomous drone to analyze and recognize the shape of trash on the beach and pick them up accordingly through various image processing algorithms such as histogram backprojection and morphological transform. - Also used computer vision and machine learning algorithms such as ORB feature descriptor with the bag of words model for feature extraction and training the model with support vector machines to classify against different types of trash. - Designed the intelligence of a robot that will compete in the Amazon Picking Challenge through computer vision, machine learning, and the use sensors. - Researched in object detection algorithms, such as correspondence grouping, in 3D space using depth maps produced by a Microsoft Kinect. - Implemented a 2D object recognition model using convolution neural networks. Currently researching in effective image segmentation algorithms for object localization. Also leading team of industrial engineering students (a senior undergraduate + master students) for the development of the autonomous robot. Worked on on designing visual and segmentation algorithms for medical images
Lehigh UniversityComputer Vision and Machine Learning Research Assistant
May. 2015 - Aug. 2015- Participated in the Lehigh SmartSpaces REU program in computer vision and machine learning. Researched in developing an emotion classifier that is robust to common image weaknesses such as lighting, unique facial expression, and unique appearance by making a facial movement based classifier. - Implemented this by using computer vision algorithms such as dense optical flow as well as machine learning algorithms such as support vector machines. - Research paper was presented at the IEEE MASS 2015 REUNS workshop held in Dallas, Texas and is now published (first author).
Intch is a Professional Networking App for the Future of WorkSee more people like Anthony on Intch
300k+ people
130+ countries
AI matching
Get Intch if you'd like to
518018 people
15
Founder & CEO @ ZeroBlind
16
Program Manager @ DISH Network
18
Senior Software Engineer @ EarnIn
118731 people
15
Founder & CEO @ ZeroBlind
42
Senior Mining Engineer @ Riot blockchain
16
CEO/Data Engineer and Consultant @ Propeltech Services LLC