Skip to content
Skip to case study
Back to work
CASE STUDY / 01LinkedInX

Security observability · 2026

CyberSentinel

A local observability hub that turns scattered system signals into clear, auditable decisions.

TL;DR

A local Python daemon that audits processes via WinAPI, correlates suspicious signals and translates them into a single panel with severity and context—nothing leaves the machine.

100% localZero telemetry sent off the machine
3 layersProcess, network and log correlated
v1 → v3Heuristic detection iterations
  • Local attack-surface audit
  • Prioritized alerts
  • Guided mitigation flows
PythonWinAPIPowerShell

01 / CONTEXT

The problem

Indicators of compromise on a Windows machine live scattered across Event Viewer, active processes and network connections. Catching an anomaly in time means crossing all of that by hand.

02 / DIRECTION

The solution

A local Python daemon that audits processes via WinAPI, correlates suspicious signals and translates them into a single panel with severity and context—nothing leaves the machine.

03 / EXECUTION

The process

01Architecture
02Prototype
03Iteration
04Delivery
  1. 01Mapping the WinAPI calls needed to read process handles and signatures without elevated privileges.
  2. 02Prototype anomaly heuristics over a dataset of our own processes.
  3. 03Iterating the heuristics to cut false positives on uncommon legitimate apps.
  4. 04Packaging as a local service with guided mitigation flows instead of bare alerts.

04 / RESULT

Impact signals

100% localZero telemetry sent off the machine
3 layersProcess, network and log correlated
v1 → v3Heuristic detection iterations

05 / IMPLEMENTATION

A decision in code

monitor.py

import ctypes
import win32process
import win32con

def audit_running_processes():
    """Scan process memory surfaces for anomalous handles."""
    print("[*] Initializing CyberSentinel Local Observability...")
    anomalies = []
    # Query system information via WinAPI
    processes = win32process.EnumProcesses()
    for pid in processes:
        try:
            h_proc = ctypes.windll.kernel32.OpenProcess(
                win32con.PROCESS_QUERY_INFORMATION | win32con.PROCESS_VM_READ,
                False, pid
            )
            if h_proc:
                # Audit handle safety signature
                sig = check_security_signature(h_proc)
                if not sig:
                    anomalies.append(pid)
                ctypes.windll.kernel32.CloseHandle(h_proc)
        except Exception:
            continue
    print(f"[+] Scan finished. Detected {len(anomalies)} process anomalies.")
    return anomalies

06 / LEARNING

Reflection

I would separate the collection layer from the heuristics earlier: mixing them at the start made iterating on false positives slower. That is the kind of debt you only see once the system is running.