Detection Engineering & Use-Case Management

Build and tune detections around real attack paths

Built over 15 detections across 160 active analytics rules and refined over 50 alerts, mainly individual instances and metadata, with tuning focused on fidelity and manageable analyst workload.

UK Water UtilityDetection development, alert tuning and detection lifecycle engineering2 min read

An alert needs to tell the SOC what attacker behaviour it may represent and why that behaviour matters. Across a bank of 160 active analytics rules, I built over 15 detections and tuned over 50 alerts, primarily individual instances and associated metadata, connecting the work to real-world threat scenarios and attack paths.

I worked on both detection logic and the context carried by the alert. Linking these back to threat scenarios and attack paths gave the engineering work a concrete investigative purpose: what activity the rule is intended to identify and how the alert relates to a possible intrusion. I reviewed legacy whitelist exceptions with detection fidelity and analyst workload in mind. I aimed to balance the SOC queue at no more than five L2 alerts per analyst per shift. This kept the focus on detection fidelity and the workload reaching an analyst across the queue.

The Use Case Pipeline took work through business justification, log prerequisites, KQL implementation, review, release, tuning and post-release testing. I introduced priorities, stages and named owners, and supported versioned Sentinel rule logic so changes had a recoverable history.

I also established a central Planner lifecycle for detection, telemetry and control gaps. Alongside that delivered work, I co-developed the wider Detection and Response framework: threat scenarios and attack paths linked to use cases, data requirements, parsers, rules and validation evidence.

The proposed Azure DevOps model separated ownership, priority and approvals from the technical content held in Git. It provides a design for tracing a business risk through the data, detection and test that address it. The delivered pipeline and versioned rules are the operational foundation; the broader linked coverage model remains the next stage.

The engineering modelFrom a threat scenario to a tested detection.
Business riskAssets · impact · threat scenariosThreat modelling ↗
Observable evidenceOn-prem / cloud · telemetry · data qualityTelemetry requirements ↗
Detection use caseRequirements · dependencies · ownership
Versioned Sentinel ruleReview & releaseTest & refine

I delivered the detection and tuning workflow and contributed to Sentinel rule versioning in Git. The wider Azure DevOps relationship model and live coverage view remain proposed extensions.

Methods, evidence & technical decisions

Start with the threat and the available evidence

I helped connect business impact and attacker behaviour to the data needed for a detection. Telemetry work included DNS source attribution and cloud/on-premises traffic context. The use case records what needs to be observed and which data, parsers and investigation context it depends on.

Open this detail ↗
Take a KQL rule through review and testing

The delivered Use Case Pipeline defines justification, source-log prerequisites, KQL implementation, logic review, release and tuning, followed by a test trigger and deeper analysis. Priorities and named owners make both new detections and tuning requests actionable. I also contributed to the repository of versioned Sentinel rule logic.

Open this detail ↗
Separate detection logic from delivery tracking

In the proposed Azure DevOps design, work items carry ownership, priority, approval and delivery status. Git holds the detailed KQL, schemas and mappings. I co-developed the ten-type work-item model as an extension to the delivered detection lifecycle and supported the versioned-rule work.

Open this detail ↗
Make telemetry and validation dependencies visible

The central Planner backlog gives detection, telemetry and control gaps an intake, triage/refinement, delivery and closure lifecycle. The proposed coverage model then links a threat scenario to data requirements, detection logic and validation evidence, so each part of the coverage claim can be assessed separately.

Open this detail ↗
Build detections and tune alert context across the rule bank

My work covered over 15 detections built and over 50 alerts tuned, primarily individual alert instances and associated metadata, within a bank of 160 active analytics rules. Tuning included reviewing legacy whitelist exceptions and connecting alert context to threat scenarios and attack paths. I aimed to balance the queue at a maximum of five L2 alerts per analyst per shift, prioritising useful signal and manageable investigation workload.

Open this detail ↗

Skills used in this work

This work connects to

Detection engineering

Your exploration

Viewed history

Saved only in this browser.

Ask my portfolio

Ask about my work.

Ask about an investigation, a technology or my contribution. Answers link to the relevant case and technical detail.

Every answer links to the work.
How this works

When AI is connected, your question, recent questions and relevant published pages are sent to MiniMax to select supporting passages. Otherwise, this searches the site directly. Answers quote the saved website and link to their sources. This website does not save chat transcripts. Clearing or leaving the page clears this conversation.

Help improve this website

Report an issue

Spotted something broken or inaccurate? Tell me what happened and what you expected.