Skip to content

dev_notes.md

Latest commit

 

History

History
45 lines (38 loc) · 2.55 KB

dev_notes.md

File metadata and controls

45 lines (38 loc) · 2.55 KB

v0.3.1 - Memory Fine Tuning Update

Fine tuning memory functions

  • Added memory context to longterm memory decision tool to prevent duplicates and enrich memories
  • Memory node uses llama3.2 to search user's memories via get_context, then uses gpt-3.5-turbo to interpret them
  • Changed memory related functions to async

Router node adjustments

  • Gave router node access to chat history
  • Switched router node to gpt-3.5-turbo for better reasoning when user refers to previous chat messages for context

    • Example - User: What tasks do I have today? Assistant: Your tasks for today are... User: What about tomorrow?

      Previously using llama3.2 without context, the router node would only see #What about tomorrow# and it would route this to the weather node. Now with chat history and gpt-3.5-turbo it can reason with previous messages and understand this needs to be sent to the task mode. _A small amount of testing has shown success, but I assume there will be hiccups with the task node recieving only the last user message and no context. I will address that on a future version.

Implemented Zapier for google calendar creation when creating new tasks

  • Zapier API with defined dictionary for information about tasks created

Chores

  • Changed [DEBUG] prints for continuity
  • Added DEEP_DEBUG for larger context debugging for things like forecast returns, internal llm prompt outputs, etc.
  • Added DEBUG and DEEP_DEBUG to backend/config.py for centralized adjustments of debugging prints
  • Added DEBUG and DEEP_DEBUG to .env file to adjust true or false

Future upgrades

  • Include Gmail monitoring to automatically create tasks via Zapier when email sent to home email address for things like:
    • Furnace filter change alerts
    • Fixd alerts for auto maintenance
    • Home insurance yearly reviews
    • Hot tub maintenance via monitoring app
    • Cold plunge maintenance via monitoring app
    • Smoke detector battery changes via home maintenance app
    • Home server maintenance and error faults via iDRAC email alerts
    • Automatically create a task to call and schedule dog grooming/healthcare via reminder emails
    • Create template for setting up any creation of a task based on received emails

Future Future upgrades

  • Implement OpenAI API with curated promting to analyze these emails and create these tasks based on:
    • Schedule availability
    • Urgency of specific task
    • Combination of like tasks for efficiency
    • Ability to change an already created tasks for room for a more important/urgent task