Harkirat Gill

Machine Learning Engineer at CorroHealth

Designed and developed 5 GenAI pipelines and integrated into AI Studio for automated large-scale processing of medical data; created ensemble AI diagnosis identification pipeline; optimized pre-existing LLM and NLP pipelines for speed; led AI/ML team for diagnosis model testing and performance optimization; developed method for automatically generating and optimizing prompts for LLM queries.

Machine Learning Engineer at ForceMetrics

Developed rapid AI insights system to interpret law enforcement field data and improve officer service quality and impact; led design & development of hybrid language models to identify key community and high-risk events from reports using fine-tuned LLMs and hybrid NLP models; optimized AI QA system for processing large volumes of documents; led LLM-based incident tagging and event extraction pipeline; cut ML development costs by 80-95% across >30 models with new model training workflows; led data team in AI testing strategy; created ML validation studio with LLM-based document highlighting; developed NLP ensemble method boosting incident topic detection accuracy by 40%.

AI Engineer (Contract) at Stenograph

Created automated job recommender system using ML and rules-based business logic; built ML training pipeline and API for inference in deployment.

Machine Learning Engineer (Contract) at HappiLabs

Designed and developed fast ML-based automated product identification & ordering system for healthcare inventory management; increased speed of product ordering pipeline by 3x using OCR and custom image CNN models; integrated with iOS mobile app and product database.

Computer Science Instructor at Benedictine University

Led and lectured for 4 undergraduate AI/ML courses with emphasis on hands-on implementation of major algorithms through course projects.

Computer Vision Research at WildMe

Developed methods for improving animal identification from photos for wildlife researchers with improved image segmentation ML models; improved accuracy by 20% with zero labeled data using a novel segmentation approach.