Use this skill when building or pitching an AI-powered customer intelligence platform for telecom operators on Huawei Cloud. TRIGGER when the user needs — AI contact center (AICC) demo architecture, telco churn prediction pipeline, ASR + LLM deployment for call analytics, executive POC strategy for telecom, data sovereignty compliance (Mexico LFPDPPP), demo design with deterministic fallback, or Huawei Cloud ECS GPU deployment for AI workloads. Also use when preparing telco executive presentations that combine live demos with strategic narrative.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
telco-call-center-ai
description
Use this skill when building or pitching an AI-powered customer intelligence platform for telecom operators on Huawei Cloud. TRIGGER when the user needs — AI contact center (AICC) demo architecture, telco churn prediction pipeline, ASR + LLM deployment for call analytics, executive POC strategy for telecom, data sovereignty compliance (Mexico LFPDPPP), demo design with deterministic fallback, or Huawei Cloud ECS GPU deployment for AI workloads. Also use when preparing telco executive presentations that combine live demos with strategic narrative.
Telco Call Center AI
Overview
This skill provides a complete, battle-tested pattern for deploying AI-powered customer intelligence demonstrations on Huawei Cloud for telecom operators. It covers the full lifecycle: architecture design, ECS GPU provisioning, ASR + LLM pipeline deployment, dashboard setup, demo narrative design, and executive POC strategy.
The skill is based on a real Huawei Cloud engagement with a major Latin American telecom operator. It captures what worked in production, not what works in documentation.
Two demos are covered:
Demo 1 — Customer Intelligence: ASR transcription → LLM analysis → churn risk scoring → real-time dashboard. Proves the operator can detect at-risk subscribers from call center interactions.
Demo 2 — Agentic Engineering: AI-powered code modernization with legacy analysis, code generation, and test generation pipelines. Proves AI can accelerate software delivery velocity.
Narrative arc: Demo 1 shows the symptom (churn detection gap). Demo 2 shows the permanent cure (engineering velocity to outrun competitors).
Quick Start
Follow this sequence for a new telco engagement:
Provision infrastructure — ECS GPU (Pi2.2xlarge.4, T4 16GB) with Docker, security groups for ports 8000/3000
Deploy Demo 1 — ASR + backend + dashboard with deterministic fallback mode
Run scripts/smoke_check.sh for health verification, check common pitfalls below
Terraform infrastructure
Use terraform_ecs_gpu/ for repeatable ECS + security group provisioning
Core Rules
R1: ASR Must Run Locally for Data Sovereignty
Telecom call recordings contain PII. In regulated markets (Mexico LFPDPPP), the ASR model must run on local infrastructure, not via cloud API. Deploy Qwen3-ASR on the ECS GPU instance itself. Only the LLM inference may call external APIs (MaaS) if properly secured.
R2: Deterministic Fallback Is Mandatory
Live demos fail. Always implement DEMO_MODE=deterministic that returns precomputed, realistic results without any model dependency. The dashboard must never show an error state during a live presentation.
R3: Two-Act Narrative, Not Feature List
Structure executive demos as a story:
Act 1 (Demo 1): "Here's your symptom — you're losing subscribers and can't see why."
Act 2 (Demo 2): "Here's the cure — AI-powered engineering that lets you outrun competitors."
Never present demos as feature showcases. Every feature ties back to a business pain point.
R4: Open with Demo, Close with POC Proposal
First 60 seconds: live demo running. Last 60 seconds: clear, low-risk POC proposal with explicit go/no-go criteria. Slides exist only as backup if demos fail.
R5: Pre-Recorded Backup Video
Record each demo at full quality before the presentation. If network, hardware, or model latency kills the live demo, switch to video within 5 seconds. The audience should not know the difference.
R6: Never Hardcode Customer Data
All scripts and examples must use placeholders: <ecs-ip>, <region>, <project-id>, <bucket>. Never include real IPs, passwords, tokens, access keys, or customer names in checked-in files.
R7: Huawei Cloud ECS SSH Is Root, Not Ubuntu
Huawei Cloud Ubuntu images default to root user for SSH keypair auth. The ubuntu user does not work. Always use ssh -i <key> root@<ip>.
R8: NEXT_PUBLIC_ Env Vars Are Build-Time
In Next.js, any NEXT_PUBLIC_* environment variable is baked into the JavaScript bundle at build time. Changing .env requires docker compose up -d --build --no-deps dashboard. A simple restart is NOT enough.
R9: API URL Must Not Include /api Suffix
The dashboard code appends /api/tasks, /api/scenarios, /health to the base URL. Setting NEXT_PUBLIC_API_URL=http://<ip>:8000/api causes double-prefix /api/api/tasks (404).
Default Deliverables
When using this skill, produce:
Infrastructure — ECS GPU instance with Docker, security groups, and DNS configured
Demo 1 backend — FastAPI with ASR integration, churn analysis scenarios, WebSocket real-time events
Demo 1 dashboard — Next.js with scenario selection, live transcript view, churn risk visualization
Demo 2 backend — FastAPI with 3 precomputed scenarios (legacy analysis, code generation, test generation)
Demo 2 dashboard — Next.js with pipeline visualization, code diff panel, test results panel
Smoke test report — Health checks for all services, WebSocket connectivity, API endpoint validation
Executive narrative — Two-act story tying demos to business outcomes
For Demo 1: ASR transcribes audio locally → Backend sends transcript to LLM via MaaS → LLM returns churn analysis → Dashboard displays in real-time via WebSocket.
For Demo 2: No GPU required. Backend serves 3 precomputed scenarios directly. Dashboard renders pipeline steps with animated progress.
Both demos support DEMO_MODE=deterministic fallback that bypasses all model dependencies.
