Wednesday 30 August 2023

Show HN: Chatmate: Discover and create chatbots https://bit.ly/3PhZwoI

Show HN: Chatmate: Discover and create chatbots Chatmate.dev allows you to easily make gpt4 based chatbots by combining multiple chat completions and/or document retrievals. You can chain different components together or run them in parallel and then use their responses in future prompts/components. To get started, create a project and then create a chat component. A single chat component is the same as one chat completion. You can add more chat or document components that use the responses of previous components in their prompts. There are some simple demos in the "discover" page. You can also publish your own chatbots and share them at the share url. The first demo is a simple document extraction. I copied a few posts from today's hackernews and fed it as a pdf. I then used the document retireval in the prompt for the final chat completion. The second demo is a teaching assistant chatbot for a data structures course. It is made up of 4 components, 1) Standalone query, which converts the user input to a standalone question (so that the document retrieval can be improved) 2) Thought generator, generates a sentiment based on the user's input (i.e. the user seems stressed about their data structures homework) 3) Document retrieval, retrieves lecture notes from over 400 documents 4) combines all the components into one system prompt for a better response. Roadmap: 1) Code components - run any code (including network calls) 2) Conditional components - uses gpt functions to decide which components to run (components are the functions) 3) Multitenancy (i.e. publish your bot at {botname}.chatmate.dev 4) Verisoning - chat with multiple versions of your bot (easy comparison of prompts) If you have any feedback, feel free to reach out at "jmiran15@jhu.edu". https://bit.ly/3OQdQ7H August 30, 2023 at 05:22AM

No comments:

Post a Comment