In build250K-member electric cooperative Confidential system, demo on request

ACIRS: 24/7 AI Customer Support

Every member inquiry gets an answer at 3 a.m., and every conversation fills the CRM by itself.

An AI customer-support system that answers members around the clock from the company knowledge base, hands off to the right department, and builds a customer record from every conversation.

ACIRS: 24/7 AI Customer Support screenshot

Designed so every inquiry gets an answer, even at 3 a.m. (in build, launching Q1 2027)

Team of one· Solo architect & engineer

No team, no agency, no hand-offs. I owned this project end to end:

  • Requirements
  • Architecture
  • AI & prompts
  • Backend & APIs
  • Infra & deployment
Stack
LLM AgentKnowledge BaseAuto-EscalationCustomer 360Omnichannel

Where AI does the work: An LLM agent answers from the knowledge base and decides when to hand off to a human department.

What was broken, and what fixed it

  1. Problem: Members ask questions at all hours, and after-hours inquiries wait until the next business day.

    Fix: An LLM agent is designed to answer around the clock from the cooperative's own knowledge base, not general web knowledge.

  2. Problem: Inquiries arrive on several channels, and each one starts from zero with no customer history.

    Fix: Web chat, social, Viber, and Telegram feed one conversation service, and every interaction is written to a Customer-360 record, so history carries across channels.

  3. Problem: Some questions need a person, and they have to reach the right department.

    Fix: Escalation rules route those conversations to the right department's queue. The rules are explicit and testable, so the agent hands off instead of guessing.

Approach

  • Answers are grounded in the company's own knowledge base, not general web knowledge.
  • Auto-escalation to the right department when a human is needed, instead of the bot guessing.
  • Every interaction feeds a Customer-360 record, so the CRM fills itself.

What it does

AI agent

  • LLM agent that answers from the knowledge base
  • Decides when to hand off to a human

Channels

  • Web chat
  • Social
  • Viber
  • Telegram

Customer 360

  • A customer record built from every interaction
  • History carried across channels

Architecture

AI

  • LLM Agent
  • Knowledge Base

Logic & APIs

  • Auto-Escalation

Infrastructure & delivery

  • Customer 360
  • Omnichannel
  • Telegram
  1. 1
    LLM agentTelegram

    An LLM agent sits behind one conversation service that web chat, social, Viber, and Telegram all feed into.

  2. 2
    Knowledge base

    Answers are retrieved from the company knowledge base, so the agent replies from the cooperative's own content.

  3. 3

    An escalation step routes conversations that need a person to the right department's queue.

  4. 4

    Each interaction is written to a Customer-360 record, so history carries across channels.

Engineering

  • Planned guardrail: the agent answers only from the knowledge base and hands off instead of guessing.
  • Escalation rules are designed to be explicit and testable, not left to the model's judgment alone.
  • Every conversation is planned to be logged against the customer record for audit and review.
  • Channel adapters are designed to be separate, so adding or changing a channel does not touch the agent.

Outcome

  • Designed so every inquiry gets an answer, even at 3 a.m.
  • In build, launching Q1 2027.