Date: Mar 2024 - Sep 2024

Roles: Data Scientist

This project was completed while I was a consultant at Melio AI, and some of the project details have been obfuscated.

Overview

Developed and deployed a multidimensional evaluation function for an MLOps Competition as well as creating an example machine translation model that participants could use for inspiration. Inference was done using KubeFlow’s Kserve.

My Responsibilities

  • Designed and implemented a weighted scoring function for evaluating translation accuracy, resource usage, efficiency, and documentation quality.
  • Deployed automated evaluation function that triggers upon solution submission.
  • Built and deployed example solutions, including a fine-tuned t5-small model with Kserve for inference.
  • Documented processes for creating and evaluating translation models.

Outcomes/Impact

  • Delivered an automated evaluation process that was used to determine competition leaderboard and rankings.
  • Provided templates for competition participants to build solutions.

Tools Used

  • Python
  • AWS
  • AWS Lambda
  • AWS DynamoDB
  • Docker
  • HuggingFace
  • TensorFlow
  • Kubeflow
  • Kserve