People outside the tech sphere usually don't know how to correctly prompt chatGPT or other LLMs to write something decent. That's the idea behind Toastful. It asks you a series of questions and kind of guides you into providing the necessary information for writing a good speech. The final result is a much better speech that you can confidently deliver in front of people.
I don't have a background in CS so learning all the new tech stuff was really fun. I experimented with fine-tuning GPT-3.5 models as well as a method inspired by RAG to make the output in a specific tone and voice by feeding it sample speeches delivered by people. For now, the RAG inspired method is working satisfactorily but I am constantly playing with the prompts. Meta announced their LLaMa 3 model as well which seems really fast so that is on my list as well.
On the web side, I used NextJS along with Tailwind and Postgres for data persistence. The entire app is hosted on Render ($7/m starter instance).
I would love to get feedback/suggestions from more seasoned devs. My aim is to make this tool a lot better and useful for people.
P.S: The final speech is behind a paywall so if you want to try it for free, you can use FREE100 to bypass the credit card requirement.