Tachio can receive feedback, reflect on it, and then propose changes to its prompts (all through the magic of pg_cron)

[n0fray]

Tachio listens for feedback on their briefings, saves the feedback into the database, then generates a reflection on the feedback and proposes improvements to the prompt based on the reflection. If the user accepts the proposal, then Tachio can implement update to the proposed version of the prompt by simply writing to the database.

• Processing of feedback into reflections and reflections into improvement proposals
• The LLM generates reflections and proposals based on the feedback received.
• The user reviews the proposals and makes a decision (accept, adjust, reject).
• If a proposal is accepted (or adjusted), then the new prompt is used for future briefings.

[n0fray]

We want to keep track of all our past briefings (of all briefing types) and we want to collect feedback from the our users about each instance of a briefing. Identifying feedback can be another reflection layer (just like identifying todos).

We will use a recurring job that reviews all the feedback, reflects on the feedback (and the briefing that the feedback is about and the prompt that generated that briefing)m, and then generates proposal for how to adjust the prompts in the future (including an explanation for why this proposed change will address the specific feedback that was received).

Tachio regularly reviews new feedback that was received about its briefings and generates self-improvement proposals for how it can improve its briefing-generating prompts, including an explanation of why this proposal is responsive to the feedback it received.