Sunday 2nd July
Over the past few weeks, our team has been working tirelessly towards completing a prototype that is ready for testing at Apex. This phase has required us to focus on several key areas, including finalizing the list of features, making necessary functional changes, and honing our software development skills through collaboration with professionals. We are approaching release as we move into the testing and publishing phase of development.
As we progressed, we worked through many iterations and features, slowly improving them and testing them. through their implementation, we were able to solidify the foundation which would be able to server our client's needs. We recognizing the importance of delivering a finished product, and actively sought the guidance of industry professionals. Their expertise and mentorship have been instrumental in informing our decision making and remaining organized. We have learned new techniques, methodologies, and best practices that have enabled us to refine our work and elevate its quality. This including incorporating tests to check our code's accuracy and how to manage our workflow with many different members.
After carefully considering and evaluating various music licensing options, our team ultimately made the decision to utilize Youtube Music for our needs. Youtube permits the utilization of its streaming service from other platforms so long as the new client does not compete with Youtube directly and adheres to a set of basic guidelines. In order to tailor the search functionality to better suit our specific requirements, we embarked on the process of reverse engineering the search prompt. Reverse engineering the search prompt involved analyzing and deconstructing the patterns behind the search results and construct a query. We were able to gain insights into how the system functions and how we could modify it to align with our objectives. Using this knowledge we were able to construct appropriate prompts which would allow us to extract meaningful results. The core of the algorithm consists of using patient data to compile a set of prompts which align with music that was likely to have been popular during their teenage years and thus the most likely to have a lasting impact on them. This meticulous approach allowed us to tailor the search prompt in a way that maximized the likelihood of discovering music that aligns with our desired criteria.
We also decided to investigate solutions such as LangChain to use large language models to aid in identification and filtering of the songs received. This would enable classification of genre and enable the creation of a more advanced recommendation algorithm that could take into account the patients engagement with different genres. In this way the application could serve an evolving set of songs to hone in on a selection of songs that are optimal for the patient.
Stay tuned for more updates!