Having worked on academic research in machine learning and AI security during my undergraduate studies and for some time afterwards, I've transitioned to industry, founding HelmGuard AI in fall 2024. This shift reflects my commitment to protecting critical institutions from the growing threat landscape accelerated by increasingly capable AI models by developing defensive applications in cybersecurity.

I recently presented HelmGuard at Palantir's developer conference, showcasing how we're bringing Agent Studio features into production through our AI Security Platform. Our solution addresses a critical challenge in enterprise security: the fragmentation of security-relevant data across wikis, spreadsheets, and other siloed systems.

HelmGuard "terraforms" enterprise security data by ingesting and unifying scattered assets—policies, standards, architecture patterns, and compliance documentation — into a structured knowledge base that powers real-time, proactive security for both humans and AI. This foundation enables our AI Security Agents to operationalize security functions throughout the enterprise, continuously learning from actual usage to keep the data layer current and deliver precise, context-aware guidance.

During my talk, I demonstrated a specific third-party risk management workflow where our platform automates the processing of security questionnaires. Similar deployments have reduced processing times by up to 90% while ensuring human involvement only where necessary. For enterprises, this capability translates to faster business operations without compromising security integrity.

The power of our approach lies in how we structure data in the ontology, provide context-specific information through function-backed retrieval, and enable AI agents to take meaningful actions. While questionnaire automation represents just one application, these same building blocks can be adapted for various security workflows — from ticket management to documentation review — delivering immediate ROI while building toward a comprehensive security platform.

A key element of our methodology is continuous evaluation and improvement. We've implemented an evaluation suite that allows even non-technical team members to review AI agent performance, identify edge cases, and refine the system iteratively. This approach ensures our solution remains relevant and effective as organizational needs evolve — especially crucial as open-source AI models become increasingly powerful, creating new attack vectors that traditional security tools aren't equipped to handle.

Our long-term vision is to evolve HelmGuard into a self-improving security "source of truth" that enables advanced use cases such as proactive threat modeling, AI-driven code remediation, and automated design reviews. As we expand in the financial services sector and beyond, we're demonstrating how AI-driven security solutions can drastically reduce human error and manual rework while maintaining compliance with complex regulatory requirements.

• HelmGuard Website
• Watch Presentation Video