Currently, few municipal or regional authorities have access to the disaggregate freight activity data needed for planning, operational decision-making, freight externality evaluation (e.g. air pollution, collision risk), or equity analysis. Due to stakeholder privacy concerns, freight data are often aggregated by geography and/or commodity, limiting direct applicability of published data for local analysis. As a result, local freight planning and analysis typically rely on one of three approaches to approximate local activity: (1) disaggregation of large national commodity flow datasets (e.g. Commodity Flow Survey and Freight Analysis Framework) using general estimates of economic activity; (2) modeling (e.g. freight trip generation, facility location, agent-based simulation, and route optimization models); or (3) direct estimation of activities using limited sensor and probe datasets, often obtained or purchased from private sector operators or commercial data providers.  Each of these approaches suffers from severe limitations such as lack of timeliness, bias, lack of representativeness, reliance on unrealistic or unverifiable assumptions, and/or inability to validate results. Machine learning-based synthetic data generation methods may offer a potential approach to overcome limitations as well as operator privacy concerns to produce realistic data for local planning.  This project represents the first phase of an expected multi-year effort to design and construct one or more synthetic last-mile freight datasets that can address existing data gaps to inform planning and operational decision-making by local transportation agencies.