AI & Agentic Security Automation

Custom AI investigation agents orchestrating multi-source telemetry collection

I designed environment-aware AI agents that let an analyst begin with one investigative request, gather evidence in parallel, correlate activity across endpoint, identity, proxy, DLP, browser, collaboration, cloud, VPN, and physical-access systems, and package the results for human review.

Technologies
Splunk Proofpoint ITM Symantec DLP Exabeam Zscaler Proxy Telemetry Python Claude OpenAI GitHub
Custom AI Investigation Agents Telemetry Architecture

Impact & Outcomes

20+ hrs Saved

Eliminated manual cross-console querying, saving over 20 hours of repetitive evidence collection per deep case.

90% Faster Reports

Compressed first-pass investigative triage and structured timeline collation from days down to minutes.

13 Data Sources

Parallel correlation across endpoint, identity, DLP, proxy, cloud, VPN, browser, and badge telemetry.

Legal + HR Ready

Defensible, timestamped, source-bound reporting formatted for immediate executive and legal review.

The Challenge

Fragmented Data Across Disparate Silos

When investigating high-risk user anomalies, analysts historically had to jump across 6+ distinct dashboards—logging into CrowdStrike, Splunk, Proofpoint ITM, Zscaler, Azure AD, and physical badge systems manually. This created severe investigative latency and increased cognitive load.

When Generic AI Doesn’t Know Your Environment

Generic AI models do not know how a company names its indexes, sourcetypes, systems, or internal data. Without that context, they can confidently produce queries that look right but fail in the real environment. I grounded the agents in the schemas, relationships, and naming conventions analysts actually use so their output could support a real investigation.

How I Built the Investigation System

01. Environment-Aware Security Knowledge Mapping

Engineered explicit semantic maps connecting enterprise identity assets, index mappings, log formats, and tool relationships. Fed structured schema context to LLMs, enabling natural-language conversion to high-precision SPL and API queries.

02. Multi-Agent Orchestration Workflow

Built a multi-agent system where a Primary Orchestrator receives an investigative hypothesis and dispatches specialized sub-agents: Endpoint Agent (CrowdStrike/Tanium), Identity Agent (AD/Auth), DLP Agent (Proofpoint), and Collaboration Agent (M365/Teams).

What I Shipped

Automated Timeline & Evidence Synthesizer

The multi-agent pipeline extracts raw events, normalizes timestamps to a unified UTC baseline, resolves user-to-device mappings, and constructs a preliminary chronological timeline for analyst review.

Collaborative AI SPL Assistant

Built an environment-aware AI agent that works with analysts in plain language to draft, explain, and refine SPL using the organization’s real indexes, sourcetypes, and field conventions.