Update: [AEA-6157] - Exceed Context Window Issue

[kieran-wilkinson-4]

Summary

Updates the bot to use Prompt Reformulation and Fixed Size Chunking

Details

1. Slack messages filter through a "Reformulation" prompt before reaching the RAG
   • Looks for questions within the user's query
   • Finds sub-questions and important details
   • Searching knowledgebase to improve search query inc. key words, abbreviations, etc
2. Use Fixed Size Chunking to reduce risk of exceeding context window
   • Set chunk sizes lets us control the size of the documents pulled back for RAG
   • Can customise size/ overlap allowing for fine-tuning, between comprehension and hallucination, at a later date.

[github-actions]

This PR is linked to a ticket in an NHS Digital JIRA Project. Here's a handy link to the ticket:

AEA-6157

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
93.5% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[bencegadanyi1-nhs]

question (non-blocking): if we are thinking about re-enabling citations in the future, would it make sense to just comment out these tests instead of removing them?

[kieran-wilkinson-4]

Ah, these were removed due to duplication.

I realised I was breaking in-line references/ links because I was creating a bullet point for each dash.
This wasn't always the case as we'd get things like `NHS - Something Something", which do not need a new line 😅

When fixing this, these became redundant, and if I removed the new line logic then they matched existing tests.