Skip to content
View poojakira's full-sized avatar
💭
open to work
💭
open to work

Highlights

  • Pro

Block or report poojakira

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
poojakira/README.md

Pooja Kiran — Security Engineering

Pooja Kiran

Security Engineer · AI & Agent Security · Application Security · Cloud IAM

Portfolio · Technical Resume · LinkedIn

I build security tooling for AI execution paths, workload identities, model supply chains, and security detection and investigation. My work focuses on practical policy enforcement, deterministic analysis, automated regression testing, and engineering evidence that can be independently inspected.

Selected engineering work

Default-deny MCP/JSON-RPC request gateway that inspects selected agent-to-tool calls before execution, with capability authorization, prompt-injection inspection, sensitive-data/exfiltration checks, anti-SSRF controls, rate limiting, tamper-evident audit records, and SIEM integration.

  • 782 passing tests · 81.93% coverage (statement) in a dated hosted Python 3.12 CI run.
  • 69 named runtime detector signatures and nine Elastic Security rules (source-inventory counts, not real-world detection rates).
  • CI evidence · Repository

Identity → Authority: AWS Agent Identity Guard

Static AWS IAM policy analysis with 25 deterministic rule IDs covering excessive permissions, role trust, privilege-escalation paths, and audit-control weaknesses. Exports text, JSON, and SARIF 2.1.0 findings.

  • 244 passed / 3 skipped in a dated CI run; skipped checks require live AWS credentials.
  • CI evidence · Repository

Artifact → Runtime: HF Model Provenance Scanner

Non-executing inspection of supported model artifacts, serialization risks, provenance, and dependency evidence.

  • 260 passed / 1 skipped · 75.67% coverage in a dated CI run.
  • 33/33 adversarial fixtures flagged (curated) in a bounded evaluation; this is fixture evidence, not a general detection guarantee.
  • CI evidence · Repository

Attack Detection Engine / Aegis HealthSec — Telemetry → Investigation

A rule-based, multi-domain security engine covering ATT&CK-oriented telemetry, offline CAN/firmware/ARM lab analysis, and a developing healthcare investigation workflow. The unmerged development branches add clinical-access detection, PostgreSQL persistence, deterministic correlation, a FastAPI investigation API, analyst audit history, and a React SOC console. The healthcare examples use synthetic access events; MIMIC-IV Demo contributes de-identified clinical context, not actual hospital security logs or authorization decisions.

  • Milestone C, October 27, 2026: 212 tests passed in an isolated Python 3.12.3 environment with real PostgreSQL integration; a Docker Compose investigation scenario was also reported as passing. See PR #48 and Milestone C report.
  • Milestone D, October 27, 2026: 280 local tests reported after authentication, tenant-isolation, container, audit-integrity and supply-chain hardening; see PR #49 and its underlying branch. Hosted CI is blocked because repository Actions are disabled.
  • FHIR R4 AuditEvent connector: PR #50 publishes an authenticated FHIR R4 AuditEvent connector for source review (single resource, Bundle, and JSON-list inputs; normalization into the existing investigation pipeline). Its coding agent reports 296 passing local tests for Milestone E, including real-PostgreSQL end-to-end coverage. The branch is unmerged and hosted CI remains blocked; this is not released or independently validated deployment capability. CloudTrail, Sysmon, auditd, and Ruby/Go/Swift/Java/.NET SAST remain planned.

Healthcare detection PR #47 · Investigation PR #48 · Hardening PR #49

Scope: Synthetic security telemetry, bounded test evidence, and controlled local/container evaluation only; no customer deployment, real-world detection-rate measurement, HIPAA certification, or production-readiness claim. The main branch still reflects the earlier offline detection engine until the stacked PRs are reviewed and merged.

More security engineering

Engineering toolkit

Languages & platforms: Python, Rust, C++, Bash, FastAPI, Docker, GitHub Actions, AWS
Security: AI/LLM security, MCP/JSON-RPC, application security, IAM and least privilege, threat modeling, model supply-chain security
Testing & detection: pytest, SARIF, CodeQL, Elastic Security, SIEM telemetry, automated regression testing

Background

  • M.S., Information Technology, Arizona State University (2026)
  • Graduate Teaching Assistant and IT Grader
  • Business & Compliance Lead, AEROSEC academic technology externship (Aug–Dec 2025)

Connect


Metrics above are tied to specific dated CI runs or bounded fixture sets. They are not claims of production deployment, customer adoption, or universal detection effectiveness. Review each repository's scope, limitations, and license before reuse.

Pinned Loading

  1. mcp-agent-security-gateway mcp-agent-security-gateway Public

    Default-deny MCP/JSON-RPC security gateway for agent tool-call authorization, prompt-injection detection, exfiltration controls, and audit telemetry.

    Python 17 5

  2. aws-agent-identity-guard aws-agent-identity-guard Public

    Static IAM guardrails for agentic AI workloads on AWS

    Python 1

  3. hf-model-provenance-scanner hf-model-provenance-scanner Public

    Scan Hugging Face model repos for provenance, impersonation, pickle-risk, and supply-chain signals.

    Python 1

  4. llm-redteam-framework llm-redteam-framework Public

    Generate adversarial prompts and evaluate an offline detector for LLM red-team experiments.

    Python 2

  5. dataset-poisoning-detector dataset-poisoning-detector Public

    Detect dataset poisoning and anomalous samples with reproducible ML security checks.

    Python

  6. adversarial-ml-lab adversarial-ml-lab Public

    FGSM/PGD/C&W adversarial robustness benchmark harness for CIFAR-10 - maps to MITRE ATLAS AML.T0043

    Python