AI Staff Training Console

Waiting for Firebase initialization.

Training Operations

Turn real merchant chats into structured learnings and approved AI knowledge.

This workspace reads your existing chat threads, lets you extract training signals, and moves approved answers into a trusted knowledge base for AI Staff.

Current source

chats

Existing merchant-user chat system

Threads chats

Messages subcollection

Learnings pending queue

Knowledge approved only

Merchants

Realtime

—

Count from Firestore merchant collection

Chat Threads

Realtime

—

Training source threads from chats

Pending Learnings

Review

—

Waiting for approval or rejection

Approved Knowledge

Trusted

—

Safe records available for AI Staff

Training Pipeline

How AI Staff learns in this app

Aligned

Select a real chat thread

Use the existing merchant-user chat conversations as source material.

Create a learning record

Extract the intent, prior answer, customer reply, and corrected final answer.

Add direct admin knowledge

Admins can type approved question-and-answer pairs directly into the knowledge base.

Approve into knowledge

Only trusted records become AI Staff knowledge for future use.

Current Focus

Training-only workspace

No live reply controls here

This app is only for training, learnings, and approved knowledge management.

Scope locked

Uses existing chat infrastructure

The training app reads from your current chats collection and messages subcollections.

Source of truth

Direct admin input supported

Knowledge can now come from manual admin-approved Q&A, not only from chat learnings.

Manual knowledge enabled

Governance

Training guardrails

Chat history as source only Enabled

Pending approval required Enabled

Manual admin knowledge Enabled

Auto-learn everything Blocked

Realtime Health

Connection state

Firebase config Checking

Auth session Waiting

Firestore Waiting

Merchant Chat Inbox

Realtime threads from chats

No chat threads loaded yet.

Conversation

Select a chat

No thread selected

Choose a thread from the left to view its actual messages.

No conversation selected yet.

Approved Knowledge Hub

Here is the final trusted knowledge that the AI can use.

This page is part of training. The Training tab creates learnings, while the Knowledge tab stores the approved answers, manual admin Q&A, and cleaned business rules that become the usable knowledge base of AI Staff.

Knowledge Role Final Approved Layer

Training Input Chat Learnings + Admin Q&A

Approved Knowledge

Trusted business answers

Showing all approved knowledge.

No approved knowledge loaded yet.

Knowledge Editor

Manual admin question and answer

Mode: Create approved knowledge

Merchant ID Merchant Name

Intent Title

Question Approved Answer

Type a direct admin-approved question and answer to add knowledge for AI Staff.

Pending Learnings

Approval queue

No pending learnings loaded yet.

How this helps the AI

Knowledge strategy

Manual knowledge

Use this for FAQs, policies, delivery rules, payment instructions, schedule answers, and standard replies that should be trusted immediately.

Approved learnings

Use pending learnings from chats when you want the AI to learn from a real customer interaction and a corrected final response.

Final output

The AI should base answers on what is approved here, not on raw chat guesses or unreviewed learnings.

Training Source

Selected chat context

No chat selected yet. Open the Chats tab and choose a conversation for training.

Latest customer message

—

Previous merchant / AI-style reply

—

Create Learning

Save a pending training record

Test Console

Prompt the AI staff

Sample Output

AI response preview

Sure — the prep time of Cordon Bleu is around 25 minutes based on the merchant menu data.

Internal Access

Firebase sign in

Collection Mapping

Current Firestore paths

Merchants merchants

Threads chats

Learnings ai_staff_learnings

Knowledge ai_staff_knowledge

This app reads from chats and saves structured learning records into ai_staff_learnings, then approved records and manual Q&A into ai_staff_knowledge.