Washington DC-Baltimore Area
Passionate about tech!
- Design and maintain end-to-end ML pipelines for the Total Loss Calculator to predict vehicle repairability using a CatBoost classifier model. Successfully engineer meta-features by using predictions from auxiliary models, improving overall model recall by 0.9% or $1.35M in cost savings. Lead production deployments for modified cut score pilots, increasing Total Loss accuracy by 4%. - Independently develop and optimize CatBoost regression models to estimate Actual Cash Value (ACV) and Salvage Price of post-accident vehicles to replace original lookups. Utilize complex Snowflake SQL queries for data extraction and transformation during model training. Improve the R² score by 4% to 93% and the RMSE by $1400 RMSE for the ACV model using iterative feature engineering. Formally evaluate performance using gradient scatter plots and pandas. - Deploy scalable model endpoints through Azure Machine Learning Studio, with an abundance of experience in Conda and Docker environment configuration, inference scripting, and job orchestration located in Azure DevOps. - Lead a successful migration of the Total Loss Calculator from Azure ML to DataRobot, requiring multi-team coordination, technical planning, and refactoring of pipelines. Initiate a central migration documentation workflow to facilitate knowledge transfers. - Maintain internal Python packages to authenticate into Snowflake and Azure resources, streamlining the team’s model data connectivity and preprocessing workflows. - Train a XGBoost classifier on underwriting documents resulting in 85% accuracy. Implement a barcode decoding pipeline as a new input feature. Additionally, encouraging best programming practices by migrating scripts running locally into Azure ML jobs. - Develop a proposal to integrate LLMs into a Document Classification project, originally using XGBoost. Engineer few-shot prompting with Llama 4 Maverick that increased 3 categories by 10% in precision. Planning to finetune LLMs using LoRA.
- Completed a Workload Automation Program that relied on Power Apps, Power BI, and Microsoft Dataverse to replace the previous excel database, and collects Workload Projection data input from over 200 full-time employees and 11 managers and generates graphs to compare current and needed labor across 30 projects and 8 Departments. - Developed a Program Plan Application using Power Apps and Microsoft Dataverse that allows Managers to create new projects with field dependencies and automatically send emails to 12 Department Heads for approval with table representation and commenting features.
Scheduled hour long tutoring sessions with students worldwide for classes up to pre-calculus, Mandarin, and programming languages.