Kush Suryavanshi

Machine learning Engineer focused on building products with extra attention to detail

Boston, MA

LinkedIn
KS

About

As a prospective Machine Learning Engineer, I hold a Master's degree in Artificial Intelligence from Northeastern University, complemented by hands-on experience in data science and artificial intelligence gained through internships at BARC India, RadicalX, and the Indian Institute of Technology. Proficient in languages such as C++, Python, and MATLAB, and skilled in frameworks like Tensorflow and Pytorch, I specialize in developing innovative solutions for anomaly detection, hand gesture recognition, and multimodal emotion recognition. Eager to contribute my expertise to the field, I am well-prepared to embark on a journey as a Machine Learning Engineer, bringing a solid foundation in AI, deep learning, and data analytics.

Work Experience

BARC India

July 2023 - August 2023

Data Science Intern

  • Conducted in-depth analysis of demographic survey data, resulting in a 20% improvement in targeting accuracy.
  • Devised and implemented innovative algorithms, reducing processing time by 30% and enhancing accuracy by 15%.
  • Collaborated with a dynamic team to transform raw data into meaningful insights for strategic planning.

RadicalX
Remote

May 2023 - June 2023

Artificial Intelligence Intern

  • Collaborated with the development team to design and implement AI-powered features and functionalities of the platform.
  • Assisted in improving and refining the AI Dev Manager powered by GPT-3, utilizing natural language processing, machine learning, and deep learning technique that drove personalized recommendations and tailored learning paths for users.

Indian Institute of Information Technology Bombay
Remote

2021 - 2021

Machine Learning Intern

  • Developed a speech-to-text system utilizing Google Cloud, Pytorch, and Wavenet to convert voice recordings into text with 95% accuracy.
  • Successfully implemented a neural voice cloning system that learned to synthesize audio, showcasing a 50% improvement in recognition rate.

Education

Northeastern University

2022 - 2024
Northeastern University
3.6

Indian Institute of Information Technology

2017 - 2021
Indian Institute of Information Technology
8.73/10

Skills

C++
Python
Tensorflow
Pytorch
scikit-Learn
OpenCV
Pandas
NumPy
Matplotlib
Seaborn
Statistics
Machine learning
Deep Learning
Data Analytics
Natural Language Processing
Computer Vision

Projects

Anomaly Detection in Chest X-ray Using Autoencoder Network

  • Developed and implemented a novel anomaly detection method using dual-distribution discrepancy, resulting in a 14.6% improvement over the AE baseline and 10.8% over the MemAE baseline on RSNA Pneumonia Detection Dataset.
  • Utilized inter- and intra-discrepancy scores to identify anomalies in CXR datasets, achieving a 4.3% improvement over AE-U baseline on the VinBigData Chest X-ray Abnormalities Detection dataset.
  • Conducted experiments on RSNA Pneumonia Detection Challenge and VinBigData Chest X-ray Abnormalities Detection, which resulted in the successful implementation of our anomaly detection method to improve performance by up to 14.6%.

Autorcncoder
Deep learning
python
Pytorch

Multimodal Emotion Recognition using Recurrent Neural Networks

  • Developed hybrid models of BiLSTM and Dialogue RNN architectures on MELD Dataset for emotion classification achieving 67% maximum accuracy with a team of three.
  • Utilized various techniques like Principal Component Analysis, signal processing, and Deep Learning to analyse multimodal data from various platforms.
  • Tested 6 different modality combinations on a 5-fold cross-validation sample set resulting in an average precision of 62%.

Multi-speaker
RNN
BiLSTM
Tensorflow

Neural Image caption generation with visual attention

  • Implemented a neural image captioning architecture, enhancing model performance by incorporating attention mechanisms for dynamic focus during caption generation, resulting in a 15% increase in caption accuracy.
  • Utilized VGG and Inception V3 feature extractors to explore the impact of different visual representations on model performance, leading to a 10% improvement in image captioning quality.
  • Achieved a BLUE-1 score of 36 and METEOR score of 10.04 by implementing architecture using VGG and Inception V3, providing a 20% faster training convergence than the Inception V3 feature extractor.

Python
Pytorch
Attention mechanis