Built a full-stack demand forecast AI system integrating Spring Boot and FastAPI. Implemented end-to-end system design including ML model API exposure and forecast history management.
Key screens and system overview. See the GitHub README for full technical details.
A full-stack AI system for demand forecasting in retail and e-commerce.
Given inputs of date, gender, age, product category, and unit price, a trained ML model infers demand quantity, displays the result in the UI, and persists it as history in PostgreSQL.
Rather than a simple model prototype, this is designed and implemented as a microservice architecture — connecting a React frontend, Spring Boot API Gateway, and FastAPI ML Service via REST.
This is a personal project built with the intent of integrating AI capabilities into business systems, covering ML model API exposure, history management, and input validation.
Handled the full cycle personally: requirements gathering, system architecture design, API design, frontend / backend / ML service implementation, PostgreSQL persistence, and Docker Compose development environment setup.
Also prepared documentation including README, API specification, architecture diagram, and AWS deployment plan.
Separated Spring Boot and FastAPI to clearly distinguish business logic from ML inference responsibilities.
By exposing the ML model as an independent API, the architecture makes it easy to swap or retrain models in the future.
Forecast results are saved to PostgreSQL and viewable with search and pagination.
Implemented history persistence with Spring Data JPA and Pageable-based pagination — going beyond a simple inference API to reflect enterprise system design principles.
Input validation is placed on both the frontend and backend, balancing UX quality with API contract robustness.
Clear separation of data flow: Frontend → API Gateway → ML Service → PostgreSQL
camelCase / snake_case conversion handled at the boundary between frontend and API contract
Forecast history persistence and Pageable-based pagination using Spring Data JPA
Architecture designed to extend ML artifact management from local to S3
Docker Compose development environment that starts all services together
Full API specifications, data flow diagrams, startup instructions, and AWS deployment plan are documented in the GitHub README. Please refer to the repository to review the complete implementation.