Computer & Data Science student at Concordia University (2028) · AI engineering & AI security 📍 Montréal, QC · 🔐 Currently exploring: agent security & evaluation
I build LLM agents and secure systems, and I care about making AI output verifiable and agents hard to hijack. I've interned across enterprise consulting, security engineering and full-stack ML, and co-founded a startup that reached $8K+ in monthly revenue.
AgentShield — a prompt-injection defense layer and reproducible benchmark for tool-using LLM agents. Three agents to attack, a 100-attack corpus, four composable safeguards (detector, spotlighting, least-privilege tools, output checker), a DistilBERT detector and an AgentDojo adapter, all on free/local models. On Qwen3 1.7B the full defense cut attack success from 11% to 0% while utility under attack rose from 44% to 66%. The DistilBERT detector failed to generalize, and that result is reported openly.
Grounded Vulnerability Triage Agent — an LLM agent that turns a dependency file into a prioritized vulnerability fix report using 6 tools (OSV, NVD, CISA KEV, EPSS, upgrade checks, schema-validated report). A grounding checker ("no source, no value") verifies every claim against fetched evidence: across 18 test manifests it found 171/171 vulnerabilities, raised 0 false alarms, and blocked the 1 fabricated value in 1,026 claims.
SustainaMars — an interactive Mars growth system built around sustainable waste recycling at Jezero Crater, with plant monitoring and AI mission planning.
- Co-Founder, Carevio (2026) — B2B pharmacy delivery platform grown to $8K+/month in 7 months. Built the iOS app, website and delivery tools (TypeScript, Supabase) and designed the routing algorithm.
- Technology & Transformation Intern, Deloitte Middle East (2026) — Business Continuity Management; designed a 5-agent agentic AI pipeline to automate Business Impact Analysis (~75% fewer analyst hours modelled on synthetic data).
- Cybersecurity Engineering Intern, IronClad Family (2026) — Built a blockchain layer and Zero Trust (NIST 800-207) permission layer for a zero-knowledge vault; scanned with OWASP ZAP and Bandit.
- Full Stack Software Engineering Intern, Micropolis Robotics (2025) — Spring Boot + Angular movie recommendation platform with a Python similarity model.
- GenAI / Agents: LLM agents, tool calling, grounding & evaluation, prompt-injection defense, n8n, Make.com
- Security: Penetration testing (OWASP Top 10, SQLi, XSS), CVSS, OWASP ZAP, Bandit, Zero Trust
- ML: Machine learning, recommendation models, data analysis (Matplotlib, Seaborn)