Deciphering technical logs

How can you use AI to explain a log without confusing clues, hypotheses and a confirmed cause?

1250 € per session, excl. VAT

In short

This half-day session of 3 hours offers a guided introduction to analysing technical logs with generative AI. Working on a fictitious incident, participants prepare an excerpt, question the AI, check its interpretations and write a provisional diagnostic note supported by identifiable lines.

Why this training?

A technical log contains a mixture of normal events, warnings and errors. A dramatic message is not necessarily the cause of the incident: reconstructing the sequence of events and linking identifiers avoids drawing a conclusion from an isolated line. AI can suggest a readable explanation, but it can also invent a meaning or fill in missing context. Asking for quotations, highlighting uncertainties and returning to the log helps build a useful summary for support without presenting an assumption as a certain diagnosis. Continuing case study: The fictitious organisation « Brévanel Réservations » operates a portal where some confirmations fail. An application log excerpt brings together successful requests, warnings and timeout errors. Participants must explain the chronology without prematurely attributing the failure to the database. Application at work: After the session, reuse the template on an authorised, anonymised excerpt, then have the note reviewed by a technical adviser. Compare the claims with the source lines and record rejected hypotheses before any further investigation.

Objectives of this training

  • Identify timestamps, severity levels and correlation identifiers.
  • Remove sensitive data before sharing an excerpt.
  • Formulate a request for an explanation with quotations from the source lines.
  • Distinguish between observations, hypotheses and missing information.
  • Write a provisional note and measure the traceability of its claims.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Junior support technicians who need to read application logs.
  • Product managers and staff in functional roles involved in assessing incidents.
  • IT staff wishing to explore AI assistance with a technical excerpt.

Prior skills

  • No prior AI knowledge required; guidance is provided for formulating the first requests.
  • You should be able to open a text file and know the concepts of a request and an application error, which are recapped at the start.

What you will work on

  1. Read before asking

    Objective: Identify usable elements in the log. Line structure, timestamp, severity, component, request identifier; limitations of an incomplete excerpt.

  2. Prepare a safe excerpt

    Objective: Select the necessary context without exposing sensitive data. Secrets, addresses and identifiers; consistent anonymisation; retaining the lines before and after an error.

  3. Obtain a verifiable explanation

    Objective: Guide the AI without asking it to guess the cause. Structured request: context, excerpt, question, expected format, quotations and uncertainties; distinction between correlation and causation.

  4. Check and hand over

    Objective: Produce a usable summary that states its limitations. Structure of an incident note; checking quotations, omissions and overstated conclusions; measuring supported claims.

What you take away

  • Prepared log excerpt and context sheet.
  • Reusable analysis request with evidence requirements.
  • Table of observations, hypotheses and required checks.
  • Provisional diagnostic note and completed quality assessment grid.

Method and assessment

  • Short teaching segments followed by immediate practice on the continuing case study.
  • Individual work followed by work in pairs, with peer review of the outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Arrange access to generative AI authorised by the organisation; no production log is required. Provide a fictitious log with numbered lines, its application context and a model answer reserved for the final check.

Observable criteria: The shared excerpt contains no secrets or personal identifiers that have not been anonymised. ; The chronology follows the timestamps and distinguishes between the requests concerned. ; Each factual claim cites a relevant line; the proportion of supported claims is calculated. ; Hypotheses are explicitly separated from facts, with no cause declared certain in the absence of evidence. ; The note proposes further checks without prescribing changes in production..

Frequently asked questions

Do you need to know how to program?

No. The reading exercise covers application events explained during the session, without writing code.

Can you use a real log?

The exercises use fictitious or anonymised cases, using only tools authorised by the organisation.

Does the AI confirm the cause of the incident?

No. It helps formulate hypotheses; confirmation requires evidence and further technical checks.

Do you make changes to the system?

No. The session stops at analysis and handover; no suggested commands are executed.

Last updated: