Echo MVP

Initial MVP for Project Echo.

Connects users with a prompt engineer buffer before sending a prompt off to chat GPT. Echo analyzes the initial prompt and responds with a JSON array containing the analyzed prompt, broken up into chunks.

Chunks

Each chunk represents a segment from the initial prompt. Each comes with a threshold that determines how beneficial to the prompt additional clarification for that chunk would be. Next is the suggestions array and heading string. The heading is used to give users insight on how to clarify for that chunk, and the suggestions array contains auto suggestions for the user to choose from. Finally, there are two objects, the adversity map, and the analyzed map. The adversity map defines how potentially adverse the chunk may be, determined by a threshold, and a string for reasoning. The analyzed map contains an estimated weight, determining that chunks weighted effect on the LLM's response, and a reasoning string for transparency and insight.

UI

Users are initially prompted as the GPT website would, however, once a users begins typing, the Echo ui pops up asking if the user would like help. Here, they can toggle echo on with the cmd+k hotkey, or the toggle button, Once they send the prompt, echo analyzes it and renders the returned JSON array into the UI. The initial propmt is shown at the top with each chunk's analysis hidden in a tooltip over the chunk. Only chunks above the set threshold (default is 0.5) will appear in the UI. Users have the ability to open the settings for echo and change their threshold, showing or hiding the different suggestions. For each chunk, users can select from the suggestions chips, or create a custom chip by typing into the text input. Once a user is satisfied with their revision, they can 'revise the prompt', which returns a revisions including their clarifications.