Requirements
The project requirements for the full semester centered on improving Throwly’s AI pricing system in a way that was measurable, practical, and useful for the client. Broadly, the team was expected to improve model performance, strengthen the supporting pipeline and platform, and document the work with clear quantitative evidence.
- Improve the accuracy of AI-generated price estimates for common marketplace and household items
- Keep AI response time fast enough for practical use in the application
- Build a validation dataset and reusable testing script for future evaluation of model performance
- Explore and compare different approaches such as prompt engineering, model selection, context strategies, and data analysis methods
- Expand support to more item categories beyond the limited set already handled by the system
- Improve the quality and efficiency of training or context data, especially for common categories such as fashion and furniture
- Add data-cleaning and safety measures to reduce bad inputs, irrelevant context, and prompt injection risks
- Improve the overall API pipeline by identifying bottlenecks, removing redundancies, and increasing reliability
- Integrate external marketplace or internet-based context sources to improve pricing quality
- Add support for pricing services, not only physical items
- Contribute improvements to the website
- Learn and apply cloud or serverless deployment tools needed for the project
- Provide quantitative results through experiments, metrics, and visualizations
- Produce a final research-style report summarizing methods, findings, comparisons, and recommendations