Deep Learning Engineer2019 - PRESENTVisionX — Moonshot Labs
Technologies: Docker, Google Cloud Platform (GCP), Keras, TensorFlow, PyTorch, Python
- Implemented an algorithm to construct lines from bounding boxes for OCR output.
- Developed text classification for large class counts of more than 1,000.
- Experimented and applied GPT-2 to unstructured OCR text to convert it to a parsable format.
- Built Flask APIs to wrap information extraction using Firestore, Cloud Storage, Cloud Tasks, and other GCP native services.
- Constructed and deployed API's docker images and deployed it to Google Cloud Run.
Machine Learning Engineer2018 - 2019PackageX
Technologies: Amazon Web Services (AWS), C#, OpenCV, AWS, Python
- Maintained a product's core algorithm and fixed bugs.
- Improved the performance and accuracy of a core functionality by implementing optimized text matching using the Aho Corasick algorithm.
- Integrated Google Cloud's NER engine for raw information extraction from package labels.
- Built a customized black-box testing framework for automated testing before deployment in production.
- Constructed from the ground up a truth-labeling tool using Python and OpenCV to help implement future machine learning projects.
- Developed a tool for analysis of production environment which helps the customer support team analyze the situation in production and respond accordingly.
- Experimented with a machine learning-based string matching algorithm to take production system performance to the next level.
Associate Machine Learning Engineer2018 - 2018VisionX
Technologies: TensorFlow, Python, C++
- Extracted and isolated a staff-line detection algorithm in music sheets from the client's existing codebase (C++).
- Wrapped multiple legacy software using command-line in a single consistent codebase of modern C++ (11).
- Used an existing EAST text detector implementation to extract lyrics from music sheets.
- Developed an algorithm to associate notes extracted from legacy code and lyrics from EAST implementation.
- Experimented with EAST training and C++ implementation to optimize code to run on older hardware.