Fine-tuning BERTopic Implementation

[Hevia]

We're already using BERTopic, but we can finetune the implementation further based off of this: https://www.pinecone.io/learn/bertopic/

[Hevia]

Scope for this work also includes:

• See if you can use the embedding itself instead of creating an index for finding the id's of similar documents
• Can inference be done through ONNX or some sort of quicker runtime

Potential further work:

• Create new issues if we can further fine tune this approach using our own sentence transformers created on the corpus