TODO.txt

TODO:
- Make server production-ready (use gunicorn with docker compose)
- do retrieval for the last three lines
- Client must get the some knowledge (name, first question) from server



DONE:
- Implement threshold for retrieval. If there is nothing relevant, use default knowledge.
- create list of profanity words and stop conversation when one of these words appears.
    - actually use https://huggingface.co/unitary/toxic-bert
- Words like good, great trigger I don't know.
- if the bot answer a different question in the same way => reset dialogue and query again.
- Implement memory for last query:
    - last query's retrieval must stay in memory
    - At least query for bot_lines[-2] as well
        - This last query should be contingent on whether there are pronouns. Or if it is a lone "WRB".
- Use language detection to forbid other languages than English.