Smart SIEM — Correlation Engine & UEBA
Turn scattered security events into actionable signals.

Context
Every information system is constantly exposed to threats: intrusion attempts, account abuse, malware, abnormal insider behaviour. To keep track of their activity, servers, workstations, network devices and applications continuously generate event logs. Taken individually, these events look harmless; taken together and placed in time, they reveal scenarios typical of an attack or unusual behaviour.
However, the volume of these logs and their spread across many sources make manual analysis ineffective. A SOC analyst can neither quickly connect signals from different devices nor single-handedly rebuild the timeline of an incident. A system is therefore needed to centralise this data, reconstruct the sequence of events and identify both known attacks and departures from normal behaviour.
IMG-SIEM-01 · CONTEXT
Smart SIEM — context
Servers, hosts and network continuously generate logs.
Problem
How can heterogeneous logs be used to detect scenarios that only emerge across several events, and to spot abnormal behaviour that no static rule describes?
Solution
Design and development of a smart SIEM platform to turn scattered events into contextualised alerts. The platform collects logs from every source, converts them into a common format and analyses them through two complementary approaches: rule-based correlation, which detects known scenarios, and behavioural analytics (UEBA), which detects departures from a user's or machine's usual activity.
IMG-SIEM-02 · SOLUTION
Smart SIEM — solution
Starting point: scattered logs, hidden deviations.
Contribution
Team project. On the back-end, I designed and developed:
- the correlation engine, which evaluates rules against normalised events (thresholds, repetitions, time windows, sequences) to link signals that mean little on their own;
- the UEBA module, which builds each user's or machine's usual profile and flags deviations, such as a login at an unusual hour or an atypical activity volume.
Features
- Logs from multiple sources are collected and mapped to a common structure.
- Events from different sources are linked by thresholds, repetitions or sequences.
- Each user or entity has a baseline profile against which its activity is compared.
- Detected signals are assessed and prioritised by risk level.
- Analysts review alerts with their context to carry out investigations.
To verify: actual status of SOAR, ticketing, reporting and some frontend views.
Architecture
An event emitted by a server or network device is first captured by the collectors, then normalised and enriched. Redis Streams distributes it to the processing stages, while Elasticsearch indexes it for search and history. The correlation engine matches it against rules alongside neighbouring events; the UEBA module compares it with the entity's usual profile. Both results feed the risk assessment, which produces the alert shown to the analyst.
IMG-SIEM-03 · ARCHITECTURE
Smart SIEM — overall architecture
Sources emit their logs.
Stack
Python, FastAPI, Redis Streams, Elasticsearch, PostgreSQL, React, Vite, Docker Compose, Nginx.
Benefits
- Scattered events are brought into a coherent, chronological view.
- Attacks that only show across several events become detectable.
- Behavioural analysis covers what explicit rules cannot anticipate.
- Analysts prioritise investigations using the context provided.
Links
This case study distinguishes the overall project scope from my personal contribution. No undocumented result or metric is presented.