Ashwin U Iyer

I am a Machine Learning Developer who also dabbles in Web Development technologies to create end-to-end solutions.

Machine Learning Intern at Tata ELXSI

  • Achieved 87% accuracy with neural network on predicting user likeability for recommendation at startup
  • Packaged recommender model using MLflow for experimentation and tracking with 98% coverage
  • Designed and built ETL pipelines from Elasticsearch for time-based data pulling with 67% improvement in efficiency
  • Created data preprocessing pipelines using Pandas and PySpark that process 1 million data points in under 3 minutes

Computer Vision Intern at Shardd Analytics

  • Conceptualized an OCR system that achieved 83% digitwise accuracy using Tensorflow and Python.
  • Served a neural network model with FastAPI with a response time of 0.087 seconds
  • Trained a CNN model that achieved 9 degree angle error in image deskewing.

Research Intern at A Star Labs

  • Built a privacy-preserving Federated environment to classify Alzheimer’s using CNNs and Siamese Networks with 46% accuracy.
  • Implemented custom optimization algorithms such as FedProx, FedNova and FedAvg for federated training with an average triplet loss of 1.2 per round.

Projects

PandaTorch

PandaTorch is a small, lightweight Python package that helps create familiar scikit-learn and pandas pipelines in PyTorch. PandaTorch currently supports both CPU and GPU training and is published on PyPI

Fake News Checker

Fake News Checker is a website that uses Natural Language Processing to profile news channels’ Twitter accounts to predict the authenticity of news with upto 80% accuracy. The model also has an API endpoint to be used in other applications.

Federated Alzheimer's Classification

Classified MRI images to understand extent of Alzheimer’s using a PyTorch CNN model. Ported the entire code to Flower platform for federated training to ensure data privacy

Music Recognition and Recommendation

A Music Recommender for audio based recommendation. Uses MFCCs for audio feature extraction.

Get in touch

I would love to collaborate on projects and am actively looking to improve my work experience. I am always looking to grow, learn and meet new people. Hit me up at one of my social media profiles or you can personally contact me at: