AI Engineer · Secure Agentic AI & Automation · Maryland Athletics
Akshat Patel
Building secure AI systems at the intersection of agentic AI and cybersecurity.
Microsoft Foundry & Copilot Studio · Vulnerability management 600+ endpoints · AWS Certified Security – Specialty · MEng Cybersecurity, UMD
About
I'm an AI Engineer with a security-first background, building staff-facing agents, governed AI workflows, and secure automation systems at Maryland Athletics. Using Microsoft Foundry and Copilot Studio, I translate departmental workflows into AI automations that return time to Marketing, HR, and Operations — while applying least-privilege design and OWASP/NIST-aligned risk thinking throughout.
Beyond the day-to-day, I built MomentumEngine — a secure agentic AI research platform on AWS Bedrock combining a deterministic ranking engine with LLM agents — and have hands-on depth in vulnerability management, cloud security, and detection engineering.
MEng Cybersecurity, University of Maryland College Park — GPA 3.66 / 4.0. Adjunct Professor for INST346 & INST464, responsible for 200+ students across two courses.
Privacy Challenges and Solutions in Implementing Searchable Encryption for Cloud Storage
ICTIS 2023 · Lecture Notes in Networks and Systems, Vol. 719 · Springer Nature
Skills
AI Engineering & Agents
Cloud & AI Infrastructure
AI Security & Governance
Security Operations
MLOps / Observability
Security Testing
Programming & Data
Enterprise Platforms
Experience
- ▸Built and deployed staff-facing Microsoft Foundry / Copilot Studio agents in Microsoft Teams, translating departmental workflows into governed AI automations with reusable instructions, controlled data handling, and user-focused interaction design.
- ▸Delivered AI-enabled workflow automation for Athletics staff, returning time to Marketing, Development, HR, Business, and Leadership by reducing repetitive manual work across link generation, campaign review, reporting, and operational information retrieval.
- ▸Built a tracking-link generator agent for Marketing that standardized UTM campaign URL naming and reduced link creation time from 5–8 minutes to 1–2 minutes per request.
- ▸Designing an AI-assisted Email Performance & Revenue Tracking workflow that ingests 2 weekly Excel exports, validates campaign data, evaluates CTR/unsubscribe/bounce-rate thresholds, and produces AP-style insights for faster campaign review.
- ▸Run AI office hours and discovery sessions across 5 departments, documenting 8+ AI use cases and converting ambiguous staff pain points into secure automation requirements and agent designs.
- ▸Presented a 3-pillar AI initiative overview to ~500 Athletics staff — covering AWS data infrastructure, operational AI tools, and academic/sports partnerships — translating technical architecture into accessible, stakeholder-ready narrative.
- ▸Lead weekly vulnerability management and remediation reviews using Rapid7 InsightVM, partnering with DIT to prioritize and track critical findings across 600+ endpoints, servers, and SaaS-connected systems.
- ▸Apply security-first agent design across all AI workflows: least-privilege access, approved data sources, user permission boundaries, prompt/output review, and risk-aware deployment planning.
- ▸Supported security alert review and case documentation by analyzing endpoint, authentication, and network activity for escalation to central IT.
- ▸Provided Tier 1/Tier 2 IT support in a fast-paced higher-ed environment, resolving most Tier 1 tickets within 1 hour across Intune and JAMF-managed systems.
- ▸Coordinated a data inventory audit across 500+ assets and 50+ staff and vendors, improving visibility into system ownership, data handling, and retention.
- ▸Documented incidents using NIST and MITRE ATT&CK terminology and delivered 4 security awareness newsletters on social engineering and typosquatting.
- ▸Assisted in marketing strategy and fan experience at UMD Athletics events across multiple sports.
- ▸Lecturer for INST346 Infrastructure, Architecture & Technologies and INST464 Decision Making for Cybersecurity.
- ▸Responsible for 200+ students, delivering lectures, labs, and case studies integrating real-world security tools and scenarios.
- ▸Designed project-based assessments to strengthen applied technical skills and critical thinking.
- ▸Assisted Dean Dr. Keith Marzullo in delivering instruction for INST 346 (Technologies, Infrastructure and Architecture).
- ▸Led lab sessions on network security and distributed systems; guided students on complex technical concepts in network architecture and data management.
- ▸Write about building secure AI systems, detection engineering, and the security decisions that don't make it into documentation — from threat modelling and cloud architecture to real incident analysis.
- ▸Cover the practical overlap between AI engineering and security: where the frameworks fall short, what actually works in production, and the tradeoffs engineers face when both domains are in play.
- ▸Engineered and deployed a Snort-based IDS/IPS on a live university network segment, writing and tuning custom rules across 7+ attack categories including DoS/DDoS, backdoor, SMTP, and application-detection rule sets.
- ▸Validated detection and prevention effectiveness through controlled attack-scenario testing, achieving 95%+ detection and blocking coverage for defined unauthorized-traffic scenarios.
- ▸Built a real-time monitoring dashboard with a mobile-friendly interface surfacing alerts, logs, and packet activity from 1M+ daily captured packets.
- ▸Identified and helped remediate 30+ critical vulnerabilities in a university e-governance platform, including misconfigurations, authentication/session weaknesses, SQL injection, XSS, and input-validation flaws.
- ▸Led a 13.5-month cloud-security research effort on privacy-preserving search over encrypted cloud data, comparing SSE, PEKS, proxy re-encryption, and multi-user searchable-encryption models.
- ▸Analyzed key leakage risks including index privacy, search-pattern leakage, access-pattern leakage, keyword-guessing attacks, and file-injection attacks.
- ▸Synthesized mitigation approaches including ORAM, PIR, secure indexing, and forward/backward privacy for privacy-preserving cloud-storage designs.
- ▸First-authored a Springer Nature conference paper published in ICTIS 2023 (LNNS Vol. 719).
Education
Bachelor of Technology in Electronics & Communication Engineering
Charotar University of Science and TechnologyJul 2019 – May 2023
- ▸First-authored Springer Nature conference paper: "Privacy Challenges and Solutions in Implementing Searchable Encryption for Cloud Storage" — published at ICTIS 2023 (LNNS Vol. 719)
- ▸Led 13.5-month cloud-security research on privacy-preserving search over encrypted cloud data with Dr. Trushit Upadhyaya and Dr. Upesh Patel
- ▸Club Coordinator at Code for Cause · Board member for university magazine · Internshala Student Partner
- ▸Ashoka House Captain
Projects
Current Focus
Building and deploying staff-facing AI agents and governed automation workflows using Microsoft Foundry and Copilot Studio at Maryland Athletics.
Leading vulnerability management across 600+ endpoints using Rapid7 InsightVM, applying NIST AI RMF and OWASP LLM Top 10 thinking to AI-risk reviews.
Developing MomentumEngine — a secure agentic AI research platform combining an 8-factor deterministic ranking engine with LLM agents on AWS Bedrock.
Pursuing the Microsoft Azure AI Engineer Associate certification.
Writing at the intersection of AI engineering and cybersecurity on Medium.
Writing
All posts ↗I Spent a Day With Claude Fable 5. Here's What Actually Changed.
Fable 5 is Mythos — Anthropic's restricted-release model — now available publicly, with the dangerous drawers locked. One messy spreadsheet, one four-stage security audit, and the one architectural decision that actually surprised me.
MCP vs Agentic AI: The Difference Almost Everyone Gets Wrong
Calling MCP 'basically agentic AI' is like calling a power strip a project manager. A clear breakdown of what each actually does and why the distinction matters for AI engineers.
How to Do Threat Modelling with STRIDE (Step by Step on a Simple Web App)
A step-by-step walkthrough of threat modelling using the STRIDE framework applied to a simple web application with browser storage, cloud backend, and OAuth.
Credentials with 2FA: How It Works and How Attackers Bypass It
Explore the techniques used by attackers to bypass Two-Factor Authentication (2FA), including a breakdown of ReelPhish and other sophisticated tools.
Cisco Duo Security Breach: Third-Party Vendor Compromised
A deep dive into the Cisco Duo security breach where a third-party telephony provider was compromised, exposing MFA SMS and VOIP data.
From Zero to Security+: My Journey, Struggles, and Success in Just One Month
My experience preparing for the CompTIA Security+ exam, overcoming challenges, and the strategies that led to my success in just one month of focused preparation.
Get In Touch
I work at the intersection of AI engineering and security. Most interesting problems live somewhere between the two.
Currently at Maryland Athletics, shipping agent workflows across Marketing, HR, and Operations. If something here resonates with what you're building, reach out.