Chat 🗣️

*Evaluated the use of metadata injection in improving word vector embeddings

*Introduced multiple enhancements to the TransE model for use on Reddit data knowledge graph completion

*Enhanced models achieved 33.85% Hits@10 accuracy (TransE baseline 31%) on prediction task

*Designed and launched end-to-end workflow management and diagnostics application with React, Cloudfront, and Cognito

*Engineered Lambda proxy + API Gateway serverless backend to interact with Step Functions and DynamoDB that handles 1000 transactions per second

*Built AWS CDK + Cloudformation continuous deployment pipeline enabling developers to automatically test and deploy service to multiple stages and regions

*Won 1st place in internal Computer Vision classification competition; researched and leveraged state-of-the-art CV strategies such as transformers, cross-validation, and ensembling

*Migrated NavBuddy matching algorithm to DistillBERT NLP model

*Trained model with semi-supervised learning and NLP data augmentation, achieving 97% accuracy and 98% ROC-AUC

*Revamped NavBuddy UI with Bootstrap 5.0

*Created ExpressJS API as an interface to interact with Google Sheets and Airtable databases with 300+ cases and resources

*Launched NavBuddy, a web app and algorithm identifying 1000+ matches between cases and resources with Heroku and Flask