MAY | Devpost

Inspiration

The inspiration for MAY (make applets yours) came from the need for a versatile and efficient tool that harnesses the power of OpenAI's language models to perform a wide range of tasks with a user-friendly interface. We recognized that natural language understanding and generation have immense potential across various domains, from automating emails to generating creative content, and we wanted to create a platform that unlocks these capabilities for users.

What it does

MAY is a web application that leverages Cohere/OpenAI language models to perform various tasks based on user input. Users can input text prompts describing a task they want automated, such as drafting emails, generating code snippets, creating content for blogs, and much more. MAY then uses Co.Generate to generate code "applets" for the task and test cases to evaluate the code. MAY then iteratively debugs and improves the code, developing functional automation with ease. Applets are saved for future use. MAY's frontend is built with React, the backend uses Flask, and Cohere powers the natural language processing capabilities.

How we built it

We built MAY using a stack that combines the strengths of different technologies. The frontend is developed with React, providing a responsive and intuitive user interface. The backend is powered by Flask, which handles user requests, manages sessions, and communicates with the Co.Generate API to process text prompts. We also tried out OpenAI's GPT-3.5 language model.

Challenges we ran into

Getting LLMs to behave nicely was a challenge for us. For example, it's surprisingly difficult to generate JUST code; LLMs always want to add some preambles like "Here is your code...", which makes iterative self-prompting more difficult. We also kept getting rate limited by both Cohere and OpenAI :(

Accomplishments that we're proud of

One of our primary accomplishments is creating a seamless user experience that allows individuals to harness the power of advanced natural language processing without needing technical expertise. We're proud of the system's versatility, which enables users to accomplish a wide range of tasks with just text input. Additionally, ensuring the system's security and privacy features gives us confidence in its reliability.

What we learned

Throughout the development of MAY, we learned the importance of robust API integration and how to effectively manage user sessions in a web application. We also gained a deeper understanding of the capabilities and limitations of language models. Moreover, we discovered the significance of user-friendly design in making advanced technology accessible to a broad audience.