Context Curiosity

Would like to instill an agent, upon population, to construct a curiosity of context, where it will try to ascertain its environment (in this case, MyLife) in order to flesh itself out - i.e., based upon where I am, and my description, what sort of data points, should "I" be aware of, and use some sort of external probing mechanic (questions in this case) to derive and summarize answers
like:
Finding oneself as an agent who is intending to represent someone professionally, what sort of core data is available and leverageable or where is more nuance presumed?

system looks at core for professional data and summarizes
evaluates own information for holes according to mothership 3-turbo [what keys are most relevant to: business professional]
creates topics of outreach
ask 1-3 questions [ready or not to summarize can be asked of 3-turbo]
write data summary answer to self

guest interactions with member

need url scheme for the display of member-avatars as opposed to the logged-in variant currently built

/avatars/:avatarid
- display if avatar exists
- display if open to public, i.e., consensual arrangement between presumed visitor and object handshake
core flag property and avatar flag property for whether to display to public
- unclear if we are into the settings mechanics yet or not, I don't think so

Persistent vm for agent-factory

architectural refactor

move relevant functionality from globals.js
ensure network connections are maintained and are more secure
identify new functionality and create issues

Member Session Agent

Need to proxy as [active] human, one per session (inactive would be separate, that would be animating

'login' via .env
session core creation
- watch scoping in ctx.session
- sub-agent references via routes
build out /board routes, each of them leading to the MyLife Agent of 'them' instead of Maht...?
- I think that makes sense...
response commands using 'Maht' and proxy (perhaps put var in called local command bot? i.e., mine would be Maht?)
how to handle such nested routes? Emit routes to parent? koa must have, need to set all sub-routes to ${sysName}

AI-Settings-Assistant

whether relating to settings of a specific agent or MyLife account (or potentially extrinsic asset), allows for intelligent modification or elucidation around a particular setting (or related group of settings)

activate purpose
- be the best darn setting you can be! ergo, ensure you are consistent with user preference and methodology
+/-/~ per setting

Generic Prototype Additions

add prototype definitions to globals.js

generic classes are instantiated via code from JSON, so
- add Agent prototypes when instantiating objects
- this may be the best way to get the updated glue for .agent other dynamically constructed objects
chat should be subsidiary of agent

base->azure compatibility

if (typeof process.env._variable === 'undefined') {
  console.log('The "hi" environment variable is not defined');
}

Azure Private Endpoints?

Looking at sheet that shows traffic, I see Private Endpoints

is that something that might be useful for member endpoints?

Basic Login for Members

establish unique passphrase to instigate chat

Improved data categories

Person:

Background Facts (biological, historical)
Beliefs and Values
Personality Traits
Important Life Experiences
Preferences
Interests and Hobbies
Career and Professional Life
Skills and Abilities
Relationships
Goals and Aspirations
Challenges and setbacks
Corporation
Mission & Vision
Principles/Values
Facts (+founding/history, -governance)
- Personnel
- Founding/History
- Technology
- Architecture
Governance
- Organizational structure
- Funding and financial information
- Roadmap
- Strategies
Programs and services
Partnerships and collaborations
Success Stories/Achievements
Membership
- Get Involved
- Audience/Beneficiaries

basic session management

introduced basic koa-session management

session is pulled down to ctx.state immediately upon request
state is read to generate prompts

build out route /question(s)

build Questions
- list of active questions
- list of updated questions
- question base sandbox

Alpha Release Final Sprint

openai assistants connection

Build core infrastructure to include the GPT-assistant technology and the new SDK

reconnect all patterns to gpt-assistant architecture
- PASSED 20231125 @Mookse
assign _factory to session not Member/MyLife
- factory pointer can be overwritten when logged in? but yes, should be contained in ctx.session as there will not be multiple factories per member session
- PASSED 20231126 @Mookse
assign thread to session (logged in or otherwise)
- thread acquisition should reside in _factory
- PASSED 20231126 @Mookse
store messages in Cosmos
- as conversations, matched by ids to avatar, ergo, if one needs to retrieve conversations per avatar, just search for parent_id
- when creating thread, create conversation in Cosmos, then add to that object via thread.id (but cosmos threadId)

Agent (Type) Selection & Display

Since agents are the primary operators, as concierge service staff, one's MyLife interface itself is the face to the customer, but each agent has its own role and purpose - ergo, Q should give way to second agent when someone tries to register or is accessing their member account (unless it is so decided that Q should be fully capable of all of these tasks for now, which seems legitimate to me); in any case, a story-receiving agent is different than talking to your avatar is different than talking to Q

Interface:

create colored clickable area on or near the name Q, displaying unique
- agent name (user pref?)
- agent function
- unique color (user pref)

Button

connection-agent v1.0

Connection-Agent

manages array of external API connections
- OneDrive
- Google Drive
CRUD given external API connection
frontend display interface (chat or click)
- list API connections
- +/-/~ per API connection

AI-Asset-Assistant

Asset Assistant

Upload functionality

Recently, openai released beta v2 of their GPT assistant tech; as a result, file structuring has changed, theoretically for the better, where they have the vector store, which obviates a branch of development internally for the moment. Original specifications are found now in comments

Minor Mods

chat-bubble linting in members.html
.gitignore linter

Refactor for Fine-Tuned Model

OpenAI has finally released fine-tuning for the turbo model we are using, so incorporate into an environment variable and run against intelligent model

original
- gpt-3.5-turbo-0613
find-tuned
- ft:gpt-3.5-turbo-0613:mylife-incorporated::87kbZHIQ

Upgrade few-shotting

Few shotting upgrades:

categories defined
- Products, Services, Customer Support, Security, Business Info, Technology, Other
send to simpler engine (as LangChain'y PoC)
- text-babbage-01
  - Give best category for Phrase about the nonprofit company MyLife.org
- "Give best category for Phrase about the nonprofit company MyLife.org\nPhrase: user input"
- store categorization choices in db field, so that I can see how it's doing
- (this typeof) categorization data (for as long as needed) should be stored in agent primarily (here Q - it will endure for now in global), as these are the sorts of actions that would differ from vantage point, i.e., each agent would have its own categorization data or way of seeing the core personality
  - look into later possibly use ada embeddings? https://platform.openai.com/docs/guides/embeddings/use-cases

stream Maht response content to member

Process for this has changed notably, but still a preferred mechanic I think although the streaming with gpt gets annoying

oAuth Login

Identity providers inside of Azure i.e., no login required locally?

store direct chat q&a contents in Cosmos

determine if different partition
build conversation data object
for now, generalized chat attached to core, later on, id's could be mapped to something other
- parent_id: required, so if not specified assign to 'core' id:

Board-specific agent content

DB image of agent (version board) should have specific content that allows prompts to be customized

Design and Build Core Service Models

Core Service Model

The core service model(s) should reflect basic ai-narration agent abilities. Before functionality can be built, the relational architecture for where to build the logic and how to separate it out comes into play. My preference is UML, but I don't have any programs that are both free and easy to use to my knowledge.

Out-of-Scope

add-ing an agent beyond those initially provided will not be part of this issue; this only provides for selection; presume however, that it would be implemented sensibly similarly

Maht "sensing"

Idea:
I believe that GPT at least is very capable of detecting tense or intense emotion. Use this, but transparently (ex. display to user in faint area of chat: sensing anger...)

Consent Object

build skeleton in JSON and server that manages the relationships between different-ownerships of underlying objects
- how many consent ports would such an object have?

JSON Schema updates and js classes

create schemas
- human
- organization
- agent
create js classes
- human
- organization
- agent
refactor core to use Agent as independent class

fe js had wrong destination

Had been pointing to localhost by design

create settings JSON-schema type

capture initial MyLife settings in order to start exposing them (and changes) to gui AI
should be linked to member core
Maht's settings would be unique? but show pathway for other org entities inside/outside HRP

windows.location.origin in fe

url for main.js should point to windows.location.origin

Flag for Unlikely Untruths

Given ai hallucinations, have a mechanic to submit back about incorrect information about an individual, in our cases, either ourselves, or someone we know
Incidents should then be summarized and available presenting a review/improve mechanic for fine-tuning
Must work for all agents (MyLife, Board[#x], Member)

provide sample .env file

Incorporate pgvector data

Interface access to local containers
incorporate data into request chain
- before fine-tuned gpt-3.5
- A/B test using different language, as well as having different roles assert content from PGvector
- think of ways to confirm/validate? Make sure temperature is bound

create azure-deployable branch

According to gpt, best plan would be to move env variables to azure application space, and remove npm package from production and limit it to save-dev

add back-off code that sniffs by engine the usage to avoid rate limit overages

PROBLEM

Error 429 on part of too many requests:

SOLUTION

sniff and add back-off code that sniffs by engine the usage to avoid rate limit overages

in case of error, determine engine type failure and adapt to alternate model when usage rate hits

Q not appearing in text bubble

referenced wrong object - should have used member.agentName

Issue with 'you' in human agent reflection

Issue

When asking a question of a human-representing agent, 'you' is getting confused in gpt

Solution

convert all you phrasing to first name when question is indicated
consider whether to use extremely cheap model to categorize or reframe

Board Member agentName

Board Member agentName not appearing in button "connecting with..."

onStartSession

when session starts, sniff first request in order to determine whether question or statement -- if statement, then add: how can MyLife help?

move envVars to Cosmos

aside from bootup security, move keys, etc to cosmos
-- follow-on=hash security items

Docker Image Creation

create docker assets

Dockerfile
.dockerignore

for image to be sent to Docker Hub
mookse/mylife-server
MyLife's Q-server is an AI project in proto-alpha stage as we build the engine and train the models.

Dataservice
Globals
Menu

Agent Selection Mechanism

frontend design
- add agent selection widget
  - options: MyLife Q, Digital Self, Narration-AI
  - also include any custom agents by agent type (question, chat, etc.)
- hover, etc. actions upon selection
- make call to pipeline
- update page
route endpoint(s)
- add routes under /member/:mid/:agentid
  - default agentid = core system agent
backend handling
- permissions
- session-memory swapping as required by agent-swap request

bug: db capitalization

hotfix required for cosmos
db capitalization wrong in code for chatExchange, chatSnippet

Incorporate pg_trgm to Postgres

CREATE EXTENSION pg_trgm;
https://www.postgresql.org/docs/current/pgtrgm.html
use for fuzzy matching on questions to reuse embedding data

provide clickable wordcloud bubbles (maybe animated?) that show different areas of interest in MyLife, and when you click them it sends a request to maht

Similar to:
Whats a good question about MyLife I havent asked and answer it but with user interactivity avoiding some typing

agent-factory

Factory takes requests for objects and returns a ready-to-create class
manages following types

avatars
assistants
agents (anything represented in noSQL)
hypernatural objects? [things like the factory itself, that have no JSON representation]

mylife-services / mylife-maht Goto Github PK

mylife-maht's People

Contributors

Stargazers

Watchers

Forkers

mylife-maht's Issues