Hojin Lee

Senior Machine Learning Engineer with more than five years of combined professional and graduate-level experience building machine learning, forecasting, optimization, and analytical decision systems. Promoted to Machine Learning Engineer II at Roadie after contributing across the full ML lifecycle, including data preparation, feature engineering, model development, evaluation, production integration, monitoring, and iterative improvement. Experienced with Python, SQL, TensorFlow, scikit-learn, AWS, Docker, Kubernetes, and CI/CD, with strong cross-functional experience translating operational requirements into production-ready workflows.

Machine Learning Engineer II at Roadie (2025-01 – Present)

• Owned and scaled production machine learning systems supporting marketplace pricing, driver-delivery matching, route and batch decisions, demand forecasting, capacity forecasting, delivery-reliability prediction, and operational planning across the machine learning model lifecycle, including data preparation, modeling, software integration, monitoring, and business evaluation.
• Advised and mentored engineers and cross-functional partners on model design, feature definitions, evaluation strategy, rollout planning, and regression analysis, helping teams apply software engineering best practices while evaluating tradeoffs among predictive performance, latency, reliability, cost, and marketplace outcomes.
• Led end-to-end analysis for model revisions by defining evaluation datasets, comparing candidate approaches, slicing performance by geography and delivery type, reviewing calibration, error analysis, and failure cases, and translating findings into feature, threshold, rollout, and monitoring changes within ML Ops workflows.
• Drove alignment with backend, platform, product, and operations teams to define interfaces between machine learning predictions and delivery workflows, including service contracts, feature dependencies, response-time expectations, fallback behavior, observability requirements, and production incident investigation.
• Strengthened ML Ops practices through repeatable training and scoring workflows, data validation, reproducible evaluation, deployment coordination, drift investigation, business-metric review, and retraining decisions for models operating under changing marketplace conditions.

Machine Learning Engineer at Roadie (2022-08 – 2024-12)

• Built and improved predictive pricing workflows using Python, SQL, Pandas, NumPy, SciPy, and scikit-learn, combining delivery distance, geography, timing, urgency, item characteristics, vehicle requirements, driver supply, historical acceptance, and market behavior to support delivery compensation and cost decisions.
• Compared regression, probabilistic, and tree-based models across markets and delivery segments using cross-validation, holdout evaluation, calibration, error slicing, and operational backtesting to identify where models generalized well and where local effects required additional features or segmentation.
• Evaluated pricing models against operational outcomes rather than offline error alone, examining relationships among driver participation, fulfillment reliability, customer cost, service urgency, and marketplace economics to support practical model selection and rollout decisions.
• Engineered labeled datasets and feature sets for driver-delivery matching using geographic proximity, route direction, vehicle and equipment compatibility, delivery size, timing, driver behavior, historical acceptance, current supply, and expected completion reliability.
• Developed classification and ranking components for candidate generation, eligibility filtering, acceptance prediction, assignment scoring, and delivery-completion risk, supporting dispatch workflows that compare multiple feasible drivers under physical and marketplace constraints.
• Applied probability calibration, threshold tuning, segment-level evaluation, historical replay, and failure-case analysis to improve acceptance and reliability scores used for repricing, reassignment, escalation, and additional driver outreach.
• Worked with software engineers on REST API integration for model outputs and internal delivery workflows, aligning prediction schemas, feature availability, response-time expectations, fallback logic, logging, and downstream use of model scores.
• Contributed to route clustering and delivery-batching systems using geospatial features, clustering methods, graph algorithms, combinatorial optimization, route-cost functions, and historical simulation to identify groups of deliveries that could be completed efficiently by compatible drivers.
• Modeled route feasibility across pickup points, destination geography, delivery windows, item characteristics, vehicle capacity, route direction, stop count, estimated duration, and service constraints, connecting predictive components with optimization logic used in operational planning.
• Analyzed route and batch performance through historical replay and segment comparison, identifying cases where geographic compactness alone was insufficient because of timing constraints, handling requirements, driver availability, or interactions among stops.
• Built market-level delivery-demand datasets and demand forecasting workflows using temporal, seasonal, calendar, geography, retail-partner, delivery-category, and item-size signals to support capacity planning, pricing, peak preparation, and operational decisions.
• Developed driver-supply and capacity forecasting models across geography, time window, vehicle category, delivery type, historical participation, and pricing conditions, connecting expected marketplace supply with projected delivery volume and fulfillment risk.
• Created regression and risk models for pickup delay, route duration, on-time completion, and delivery reliability using distance, geography, time of day, stop count, driver history, item characteristics, pickup complexity, and route conditions.
• Compared statistical time-series methods, tree-based regressors, segment-level forecasts, quantile prediction, and scenario analysis, emphasizing forecast stability and decision usefulness during seasonal peaks and unusual demand shifts.
• Built reusable Python and SQL ETL pipelines supporting the machine learning model lifecycle, including extraction, joining, cleaning, data validation, feature engineering, training datasets, batch processing, model scoring, prediction persistence, and historical backfills across multiple marketplace ML use cases.
• Added data-quality checks for completeness, consistency, schema changes, unexpected distributions, and feature availability, reducing the risk that upstream data issues would silently affect model training or production predictions.
• Supported batch and near-real-time prediction workflows across AWS and Azure environments, coordinating model artifacts, Docker dependencies, Kubernetes deployment requirements, CI/CD checks, service interfaces, logging, and operational fallback behavior.
• Defined model evaluation and monitoring across accuracy, calibration, geographic stability, data quality, driver acceptance, delivery completion, route efficiency, marketplace cost, and drift analysis, supporting controlled releases and investigation of regressions throughout the ML Ops lifecycle.

Senior Business Analyst at Happy Money (2021-02 – 2022-08)

• Built SQL and Snowflake datasets for the consumer-lending funnel, analyzing application starts, completion, credit decisions, offer presentation, offer acceptance, funding, and customer drop-off across acquisition channels, product cohorts, borrower segments, and time periods.
• Developed recurring analyses of credit performance, delinquency, repayment behavior, borrower characteristics, loan cohorts, and partner portfolio outcomes, helping product, risk, operations, and finance teams distinguish short-term conversion movement from longer-term credit-quality changes.
• Applied Python or R statistical workflows, cohort analysis, regression, segmentation, statistical testing, and hypothesis testing to investigate borrower behavior, application conversion, portfolio trends, and differences among acquisition sources and credit-union funding partners.
• Created borrower segmentation analyses using financial profile, credit characteristics, debt burden, application behavior, offer response, repayment patterns, and customer goals to support product targeting, operational prioritization, and portfolio interpretation.
• Automated recurring partner reporting and data-reconciliation workflows using SQL transformations, Snowflake, Tableau, and standardized metric definitions, improving consistency across origination, portfolio, and performance reviews.
• Built and maintained dashboards for lending volume, conversion, approval, funding, portfolio performance, and partner outcomes, pairing summary metrics with drill-down analyses that allowed stakeholders to investigate meaningful changes by cohort and segment.
• Collaborated with product managers, credit and risk teams, finance, operations, and data partners to translate business questions into measurable definitions, validated datasets, analytical methods, and decision-ready recommendations.

Quantitative Analyst at TD (2020 – 2020)

• Supported quantitative analysis of financial datasets using statistical methods, data validation, reproducible workflows, and model-oriented review, strengthening experience with high-consequence data, analytical controls, and communication of assumptions and limitations.
• Prepared, reconciled, and interpreted datasets for quantitative analysis while completing graduate data science training, applying structured problem solving to financial variables, time-dependent behavior, model inputs, and result validation.

Graduate Student Researcher at Columbia University (2019-04 – 2020)

• Co-developed an NLP and machine learning research system that analyzed a large corpus of online news to study relationships between media language and sustained peace, covering data engineering, text preprocessing, exploratory analysis, lexical methods, hypothesis testing, and text classification.
• Worked within a project stack that included Python, R, Pandas, NumPy, scikit-learn, TensorFlow, BERT, Word2Vec, Doc2Vec, logistic regression, random forests, sampling, model comparison, feature importance, Matplotlib, Seaborn, and Jupyter Notebook.
• Contributed to reproducible collaborative development using Git, GitHub, Python modules, reusable utilities, unit testing, coverage checks, pylint, autopep8, R Markdown, and structured project organization for preprocessing, modeling, visualization, and evaluation.

Manager at Global Lending Services (2017-09 – 2019-04)

• Managed analytical work supporting auto-loan application processing, automated underwriting, pricing decisions, dealer operations, and portfolio performance, coordinating priorities and translating model and business findings for credit, finance, operations, and leadership stakeholders.
• Reviewed analyst output, metric definitions, data extraction, scenario analysis, and recurring reporting, improving consistency in how teams evaluated approval behavior, loan structure, dealer performance, delinquency, portfolio trends, and operational results.
• Led requirements gathering and analytical planning for cross-functional questions, balancing the need for timely business decisions with appropriate validation, segmentation, statistical interpretation, and clear communication of uncertainty and limitations.
• Standardized recurring analytical workflows and team review practices, helping analysts move from one-off requests toward repeatable datasets, documented assumptions, reconciled metrics, and decision-ready reporting.

Business Analyst at Global Lending Services (2016 – 2017-09)

• Analyzed auto-loan applications, automated underwriting decisions, borrower risk, credit attributes, vehicle information, dealer behavior, and portfolio outcomes using SQL, Excel, statistical analysis, and Python or R modeling workflows.
• Evaluated pricing and loan-structuring scenarios across interest rate, term, down payment, vehicle value, borrower risk, affordability, approval rules, and expected portfolio behavior, supporting real-time dealer decisioning and credit-policy analysis.
• Built dealer-performance analyses covering application volume, approval rates, conversion, funding activity, geographic patterns, loan characteristics, and portfolio quality, helping business leaders identify meaningful differences among dealer segments.
• Developed portfolio-monitoring and credit-performance analyses using vintage curves, delinquency trends, cohort comparisons, regression, decision trees, time-series methods, simulation, and stress scenarios to understand changes in expected loss and borrower behavior.
• Automated portions of recurring reporting, data extraction, validation, and dashboard preparation, improving access to consistent application, decision, dealer, and portfolio metrics for operational and executive reviews.

Data Analyst at UPS (2014 – 2015)

• Led a backfill-optimization analysis to identify opportunities for using otherwise empty truck capacity, applying network optimization across hub-to-hub flows, route combinations, operating costs, route-cut scenarios, and network constraints to support transportation-efficiency decisions.
• Analyzed peak-season uncertainty in package volume, vehicle capacity, hub flow, and driver availability using historical operational data and scenario analysis, helping transportation teams understand potential capacity constraints, service risks, and cost effects.
• Reconstructed part of an operational data-gathering process from Microsoft Access to T-SQL and Microsoft SQL Server, replacing manual steps with more maintainable queries, relational structures, and repeatable processing logic.
• Produced decision support connecting route configuration, capacity utilization, transportation cost, package load variability, and driver supply, establishing an early technical foundation in logistics optimization, forecasting, data engineering, and operational analytics.

Senior Design Team Member at Manheim (2015 – 2015)

• Worked on an industrial engineering senior design engagement in the automotive marketplace domain, applying process analysis, quantitative problem solving, and stakeholder communication to an operational business problem.

Undergraduate Researcher at Georgia Institute of Technology (2015 – 2015)

• Analyzed crime patterns around Georgia Tech and the greater Atlanta area using regression, time-series analysis, triple exponential smoothing, clustering, optimization, geographic visualization, and scenario analysis.
• Designed proposed police patrol zones using an integer programming maximal-coverage model, geospatial optimization, and Voronoi-based geographic partitioning, translating historical incident data and coverage constraints into an operational resource-allocation design.
• Presented preliminary findings and patrol-zone recommendations to the Georgia Tech Police Department and Atlanta Police Department, explaining modeling assumptions, geographic tradeoffs, forecast interpretation, and practical implementation considerations.

Master of Science in Data Science – Columbia University (2020 – 2020)

Bachelor of Science in Industrial and Systems Engineering – Georgia Institute of Technology (2015 – 2015)