37 lines
1.7 KiB
Markdown
37 lines
1.7 KiB
Markdown
### How it works
|
|
|
|
1. It's similar to ChatGPT, but simpler. User can sign in, sign out and store conversations with LLM model
|
|
Q: 3rd-party auth provider (e.g. google auth) ?
|
|
|
|
2. Important: It should be able to use Azure API for GPT, because next week OpenAI closes access to Hong Kong.
|
|
|
|
3. You can have a conversation like in ChatGPT, you can type it or dictate it to microphone, it will be sent to Whisper API (Azure supports it)
|
|
4. Answer will be received from API, and attached with audio that will be generated by eleven labs API.
|
|
It needs to detect the language, so if the answer is in Spanish, then audio will be Spanish.
|
|
If Japanese - japanese audio, etc. you can continue the conversation, so it has a context of this conversation for next answer
|
|
5. Audio is looped in browser by default using simple HTML <loop>.
|
|
|
|
Remarks: LangChain can be use but the answer from ai-api will be provided as is for below api
|
|
|
|
- azure api,
|
|
- whisper api,
|
|
- eleven labs api
|
|
|
|
6. Answers are stored in server - and audios too, and can be obtained via rest api, and also from this simple web ui
|
|
Remarks: ChatGPT ui alike will be provided
|
|
Q: ui example for audios part ? for rest-api, any server provided (e.g storage) ?
|
|
|
|
7. You can use any server framework like node.js, same for frontend - react for instance
|
|
Q: will your side cover related azure and api fee in develop/bug-fix stage ?
|
|
|
|
### Details needed from a contractor
|
|
|
|
1. Name and surname
|
|
2. HKID number
|
|
Q: only first 4 digit of HKID number (eng including) provided is that OK ?
|
|
|
|
Q: 唐突一問你呢份係功課黎?!
|
|
|
|
what is the level of completion of your project (POC) ?
|
|
any chance if i leave you my company contact instead of HKID ?
|