Kaushik Saha
Security architect · Azure and AI
Kolkata, India
I design security for Azure and build with AI. That covers posture and policy, detection and response on Microsoft Sentinel and Defender XDR, and the controls around AI systems and agents. 12 years in cybersecurity.
- SC-100Microsoft Cybersecurity Architect Expert
- AZ-500Azure Security Engineer Associate
What I do
Cloud security architecture
I design security for Azure estates: posture in Defender for Cloud, guardrails written as Azure Policy, and identity and network boundaries, reviewed against the Microsoft Cloud Security Benchmark.
Detection and response
I write KQL detections mapped to MITRE ATT&CK, build detection-as-code pipelines that validate a rule before it reaches production, and automate enrichment and containment with Logic Apps.
Agentic AI for security operations
I build LangGraph workflows with specialist LLM agents. They stop at approval gates, where a person decides, and the runs are traced with Langfuse.
Securing AI systems
I set prompt-injection boundaries, tool permissions and agent governance, mapped to the OWASP LLM Top 10 and the NIST AI Risk Management Framework.
Things I have built
VeriKQLLive and free
Checks Microsoft Sentinel and Defender XDR KQL with Microsoft's own parser, inside Claude or any MCP client.
sentinel-triage-agentPublic on GitHub
A LangGraph agent that triages a Microsoft Sentinel incident and waits for a person's approval before anything is dispatched.
azure-securityPublic on GitHub
A KQL learning path and a Sentinel table library with sample logs and MITRE-mapped hunts.
Cyber OrchestratorPrivate, not published
A governed multi-agent framework for security delivery, with approval gates and an audit journal.
Architecture compilerPrivate, not published
A Python engine that turns a validated architecture model into editable enterprise diagrams, with an independent reviewer agent.
Research and guides
- Governed agentic AI for cybersecurity
- Langfuse for cybersecurity: an SOC lens
- Detection-as-Code: the pipeline and governance
- Validating Sentinel detections in CI/CD
29 published topics, grouped into 6 learning paths.