Applied AI to Telecommunications Networks Training Course

Module 1: Fundamentals, data, prompts, and opportunities

Contents and applied examples

• AI Principles: AI, Machine Learning, generative AI, and AIOps explained through real-world tasks, avoiding unnecessary jargon.

• Technology Limits: What AI can and cannot do: drafting is not verifying; summarizing is not deciding.

• Information Sources: Documents, regulations, forms, tickets, KPIs, alarms, logs, tables, and operational context.

• Data Sensitivity and Quality: Public, synthetic, internal, personal, or confidential information; origin, date, context, and completeness.

• Capability Selection by Task: Classification, prediction, anomalies, clustering, generation, and multimodality.

• Anatomy of a Controlled Prompt: Role, context, task, constraints, format, and validation.

• Tool Ecosystem: Local, cloud, and corporate options: utility, limits, privacy, permissions, and selection criteria.

Practical exercises and demonstrations

• Data Traffic Light: Quick cases to decide what information can be entered into an AI tool.

• Synthetic Incident Analysis: Working on tickets, KPIs, and alarms to generate a technical report, executive summary, and critical review.

• Guided Cases: Using documents, tables, screenshots, and simulated data to evaluate technique, risk, and validation responsibility.

• AI Opportunity Map: Identification and prioritization of a specific job task.

Module 2: Application to work, incidents, communication, and automation

Contents and applied examples

• Environment Comparison: Practical analysis of local AI, cloud AI, and authorized tools, without seeking a universal winner.

• Response Evaluation: Criteria for comparing outputs based on clarity, prudence, evidence, traceability, and utility.

• Operational Analysis: Processing KPIs, alarms, and logs to identify patterns, context, hypotheses, missing data, and safe actions.

• Audience Adaptation: Adjusting the same information for a technical report, executive summary, and escalation message.

• Error Management: Plausible hallucinations, omissions, and the appearance of certainty; signals to stop, correct, or verify.

• Advanced Prompt Engineering: Insufficient prompts versus controlled prompts, and analysis of the effect of directed or biased questions.

• Automation Traffic Light: Green, amber, and red tasks; understanding that recommending is not the same as executing.

• Functional Design of Assistants: Defining permitted inputs, expected outputs, prohibited actions, and human escalation.

Practical exercises and demonstrations

• Model Benchmarking: Comparing the same prompt and case across different approaches and tools.

• Practical Case SITE-MAD-014: Ordering evidence and preparing hypotheses using synthetic data without asserting unproven causes.

• Quality Control: Human correction of AI outputs using a checklist before sending or reusing them.

• Multi-channel Communication: Creating three outputs for different audiences from the same base information.

• Defensive Redesign: Analyzing a hallucinated response and adjusting the prompt to mitigate risks.

• Design Card: Creating the practical specification of an incident assistant with limits and supervision.

Module 3: Verification, agents, prompt-injection, and governance

Contents and applied examples

• Content Audit: Verification of texts generated or assisted by AI (assuming there is no infallible authorship detector).

• Informational Rigor: Differentiating between fact, inference, opinion, and decision; validating sources, dates, and versions.

• Content Governance: Audit table and reuse traffic light (use, correct, block, or escalate).

• AI Agents in Practice: Reading, planning, editing, and tool usage capabilities within an authorized scope.

• Secure Execution Environments: Minimum permissions, working folder, change review (diff), dismissal, and rollback.

• AI Cybersecurity: Direct, indirect, and persistent prompt-injection; treating documents and external instructions as untrusted content.

• Regulatory Framework: Applied compliance with the European AI Act, GDPR, LOPDGDD, ENS, and AEPD guidelines.

• Corporate Usage Protocol: Defining permitted objectives, authorized data, approved tools, constraints, human review, logging, and escalation.

Practical exercises and demonstrations

• Text Audit: Manual and assisted review of a suspicious summary based on a synthetic case.

• Evidence Reformulation: Extracting verifiable data and rewriting assertions into objective and defensible phrases.

• Agent Simulation: Demonstrating an agent in a secure folder (inventory, plan, controlled editing, diff, and rollback).

• Defensive Security: Detecting an embedded malicious instruction and building a defensive prompt.

• Governance Application: Developing the matrix and safe usage protocol applied to a real job case.

• Closing Evaluation: Brief final form on the competencies and concepts practiced during the course.