vgebs/thinkly.ai-backend

thinkly.ai-backend's Issues

Remove all functions that are not being used from the project

Remove all of the functions we are not using in the backend.

Adjust top_p & temp depending on type of query

Adjusting the temperature and top_p parameters can change the output of the model in various ways. Here are some scenarios where you might want to alter them:

Temperature:
Creative Writing: If you're using the model for creative writing or brainstorming and you want diverse, unexpected responses, a higher temperature like 0.8 might be appropriate. This introduces more randomness in the outputs.
Technical Writing: If you're generating technical or professional content where consistency and focus are crucial, a lower temperature like 0.2 might be better. This makes the output more deterministic and focused.

Top_p (Nucleus Sampling):
High-Stakes Decision Making: If you're using the model to help with high-stakes decisions where accuracy is critical, a lower top_p value like 0.1 could be useful. This restricts the model's responses to the most probable tokens.
Exploratory Conversations: If you're having an exploratory conversation and want a good balance between diversity and relevance, a higher top_p value like 0.9 could be suitable. This allows the model to consider a broader range of token possibilities.

Remember that both parameters control the trade-off between diversity and determinism in the model's responses. They should be adjusted according to the specific requirements of your task. As the documentation suggests, it's generally recommended to adjust either temperature or top_p, not both, to avoid confusion and unexpected behavior.

Tokenize

We need to count the number of tokens used by the user before prompting.

Use this library: tiktoken

Resources:
https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
https://github.com/openai/tiktoken/blob/main/README.md

Assignments: Take average from graded assignments

Run the grading scheme 3 times in parallel, take the outputs from each, plug them into another prompt and get GPT to give the average score.

Test AWS Chalice

Check to see how AWS chalice works then get the backend running with lambda.

Stream Responses

Stream the responses for specific prompts.

ie, when user is questioning notes + assignments.

Find other cases that would benefit from streaming.

Generate the whole curriculum at once

As of right now, fetching each unit independently is not working properly, so we need to be able to fetch the entire curriculum all at once.

Additionally, having the learning objectives, textbooks, prerequisites and course name and description are seeming to confuse gpt.

So make sure the function takes in only a user prompt.

Add chat streaming capability for GPT API.

There will be a section of the notes page where the user can ask questions regarding the notes.

For this feature we will need the chat streaming capability.

Here is the streaming documentation:
https://github.com/openai/openai-cookbook/blob/main/examples/How_to_stream_completions.ipynb

We will also need a socket set up using Socket.io python client. Here is the documentation:
https://python-socketio.readthedocs.io/en/latest/

• Get the streaming capability working.
• Get the socket working.
• Get both working together.

Generate subtopics for unit

Create a function that generates subtopics for a given unit.

Accept user prompt for self learning

Allow the user to enter a prompt on what they wish to learn.

From there GPT will auto generate the learning objectives.

Backend: User asks question about text

Make a function that takes in the existing text from the notes and the users question.

Get GPT to respond to the question.

Token Count for billing

We need to be able to count the amount the tokens each user uses.
We will push each usage to firestore.

• Setup firestore on the backend
• Add a function to push the usage to the db
• For each backend call to GPT, we need to push the token count to the DB