CS50xFinalProject

My CS50x's Final Project

Requires LM Studio or other softwares that allow local LLMs to be run.

Current features:

• Converts bulleted points of journal entries accurately to a story-like paragraph.
• Ability to regenerate content with or without feedback.
• If no feedback is provided, the response would become increasingly more creative and varied.
• If feedback is provided, generated content would modify the previous output with minimal changes.
• For regenerated content, changes and new additions are highlighted, users can select which version they would like by clicking on the other version they want to remove.
• Stores journal in a database with thumbs up / thumbs down button for good or bad entries.
• Friendly UI / UX features - User inputs are updated, copy to clipboard button etc.

To run the website.

1. In VSCode, install requirements from requirements.txt with pip install.
2. Install & set up LM Studio or equilvalent with an LLM model of your choice (doesn't really work with reasoning models, I used Meta Llama 3.0 7B Instruct).
3. Run the LLM server on LM Studio or equilvalent and modify the LLM server URL in helper.py, under send_to_llm function, as well as the model used under the payload section of the same function.
4. If desired, modify the system_prompt.txt and change it as accordingly to fit your style and what you expect back from the LLM. Recommended parts to edit is personal tone, style guide and examples. Tempering with other prompts may lead to undesirable outcomes.
5. In VS Code, run python app.py and go to the localhost website the website is running on.
6. Input your points and generate content with LLM.
7. Classify the response with thumbs up or thumbs down button to save it into results.db, which can be viewed with any software that access db files.

Upcoming features:

• Further fine tuning of the LLM output by learning from the positive and negative outputs saved in results.db
• Different method of input, ie to continously input points throughout the day, and website will automatically summarise it at the end of the day OR upon user prompt.