PORTFOLIO
AI & Agent Security · Application Security · Cloud IAM Security · Model Supply-Chain Security
I secure the trust boundaries where AI systems gain authority and take action: agent-to-tool execution, identity-to-permission, and model-artifact-to-runtime.
How I work
I work at the boundaries where AI agents, identities, APIs, and model artifacts become security decisions, then back those decisions with tests, CI evidence, telemetry, and reproducible validation.
I also think about the problem being solved, how the result is validated, how people will evaluate it, and what makes the work practical to adopt. In AEROSEC, that included business and compliance analysis plus a five-year financial model presented to ASU and Honeywell stakeholders.
Three security trust boundaries
Selected build 01
Agent tool calls cross an execution boundary where authorization, content risk, egress policy, rate limits, and audit evidence need to be enforced before downstream execution.
Candidate-reported development period: Oct. 2025 - Sep. 2026. Public GitHub history begins Jul. 2026; the cited verification reflects later 2026 repository snapshots, not necessarily the latest commit.
The evidence supports implemented controls and bounded test results. It does not claim customer deployment, enterprise adoption, or universal prevention.
Selected build 02
Agent and workload identities can become over-privileged through wildcard grants, iam:PassRole, sts:AssumeRole, weak trust policies, privilege-escalation paths, and missing permission boundaries.
Candidate-reported development period: Apr. 2025 - Sep. 2025. Public GitHub history begins Aug. 2026; the cited verification reflects later 2026 repository snapshots, not necessarily the latest commit.
The evidence supports deterministic static-policy analysis. It does not claim complete effective-permission evaluation, production latency, or runtime authorization guarantees.
Selected build 03
Untrusted model repositories and artifacts can carry serialization, loader, dependency, provenance, and obfuscation risk before a model is loaded.
Candidate-reported development period: Nov. 2024 - Mar. 2025. Public GitHub history begins Jul. 2026; the cited verification reflects later 2026 repository snapshots, not necessarily the latest commit.
The fixture results are bounded regression evidence; the benign result covers only four committed samples. Neither is a universal detection or false-positive rate.
Application security case study
A white-box security review of the gateway's public attack surface, focused on authentication, authorization, parser abuse, SSRF/egress, availability controls, and evidence generation.
Protected inspection and metrics endpoints require configured API credentials; health/readiness remain orchestration-safe.
Default-deny capability and server policy gates agent tool calls before downstream execution.
Strict JSON-RPC and HTTP parsing rejects malformed requests, duplicate JSON keys, ambiguous framing, oversized headers, and unsupported methods.
Policy checks deny private, loopback, and link-local destinations unless explicitly allowlisted.
Payload limits, rate limiting, circuit breaking, and fail-closed error handling constrain abusive or degraded paths.
Tamper-evident audit records, ECS telemetry, Elastic rules, and SIEM tests make blocked activity reviewable.
Authorized review of my own application using source inspection and automated repository tests. This is not presented as an independent third-party penetration test or production deployment assessment.
How I think
Whether I am thinking as an engineer or as a builder, I keep the same spine: identity, authority, execution, telemetry, and evidence.
Professional & applied experience
Self-Directed Research
Honeywell Aerospace Technologies × Arizona State University (academic externship)
Ira A. Fulton Schools of Engineering, Arizona State University
Capabilities
Education & credentials
M.S., Information Technology (Security)
Tempe, AZ, USAGPA: 3.87/4.00B.Tech., Computer Science & Engineering
Bengaluru, KA, IndiaCGPA: 8.44/10Academic & AI participation
Arizona State University · Principled Innovation-Infused Learning Engineering
Participated in an ASU Principled Innovation-infused Learning Engineering initiative involving a Generative AI learning assistant designed to support personalized education.
Participation scope: Educational participation; separate from professional employment, internship, and research-service appointments.
Recruiting & collaboration
I'm interested in U.S.-based Security Engineering roles spanning AI and agent security, Application Security, Product Security, Cloud IAM Security, and Model Supply-Chain Security.