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.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. 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 Bharadwaj

AI Security Engineer

Agentic AI security | LLM/RAG security | MCP and tool security | IAM and least privilege | Model supply-chain security

AI becomes a different security problem when it can act. My work focuses on securing the boundaries between AI agents, tools, identities, data, and model artifacts. I build and evaluate open-source security tooling across agentic AI security, MCP and tool access, IAM, model supply-chain security, adversarial ML, and model privacy.

Recent work includes an inline MCP security gateway, an AWS IAM policy analyzer for agent roles, and a pre-load security scanner for Hugging Face model artifacts. Broader research includes adversarial robustness, membership inference, model extraction, differential privacy, prompt-injection testing, and RAG security evaluation.

I care about security engineering that is testable, reproducible, and honest about its limitations.

LinkedIn | Portfolio | Email


Current Focus

  • Agentic AI security: MCP/tool-call inspection, policy enforcement, audit trails, tool access boundaries, and data-exposure controls.
  • IAM and least privilege for AI agents: trust relationships, delegated access, risky permissions, and cloud identity review.
  • Model supply-chain security: Hugging Face artifact scanning, unsafe deserialization signals, metadata/provenance review, and pre-load risk checks.
  • LLM/RAG security: prompt-injection testing, leakage checks, RAG/tool misuse scenarios, and detector evaluation.
  • ML security research: adversarial robustness, dataset poisoning, membership inference, model extraction, differential privacy, and model privacy evaluation.
  • Engineering stack: Python, FastAPI, PyTorch, scikit-learn, Docker, GitHub Actions, AWS IAM, SARIF, JSON evidence, and technical documentation.

Selected GitHub Projects

Project Security boundary What it demonstrates
mcp-agent-security-gateway Agent to tool Inline MCP/tool-call inspection for prompt injection, PII leakage, shadow servers, exfiltration patterns, policy decisions, and audit logging.
aws-agent-identity-guard Agent to cloud identity Static IAM analysis for agent roles, risky permissions, trust relationships, and least-privilege violations.
hf-model-provenance-scanner ML system to model artifact Pre-load Hugging Face scanner for provenance, impersonation, pickle risk, suspicious metadata, and model supply-chain signals.
llm-redteam-framework LLM to untrusted input Adversarial prompt generation and offline detector evaluation for LLM red-team experiments.
model-privacy-attacks Model output to privacy risk Membership-inference and model-privacy attack evaluation for ML classifiers.
adversarial-ml-lab Model robustness FGSM, PGD, and C&W adversarial robustness benchmark harness for CIFAR-10 research experiments.
dataset-poisoning-detector Training data to model behavior Dataset poisoning and anomalous-sample detection with reproducible ML security checks.
unified-ml-security-platform Security services to platform Integration workspace for ML security services with CI, compose validation, scans, and health checks.
ml-security-command-center Portfolio evidence Static inventory of sibling repository revisions, source observations, and evidence availability.
mlsec-dashboards Evidence to dashboard Dashboard hub with authenticated API endpoints that aggregate JSON evidence from sibling ML security repositories.
mlsec-benchmark-suite Tools to fixtures Smoke-test and evidence infrastructure using small, versioned fixtures and structured JSON results.
attack-v19-core ATT&CK data to application logic Importable Python package for MITRE ATT&CK v19 Enterprise, Mobile, and ICS data models with migration-aware lookups.
PulseNet-RUL-Forecasting Telemetry to ML pipeline NASA C-MAPSS remaining-useful-life forecasting work with adversarial sensor-input checks and secure MLOps controls.

Experience And Background

  • Independent AI Security Researcher & Engineer, Self-Directed Research, Tempe, AZ, Aug 2024 - Present.
  • Business Compliance Lead & Market/Cost Analyst, AEROSEC, Arizona State University Technology Innovation Lab, Aug 2025 - Dec 2025.
  • Graduate Teaching Assistant / IT Grader, Ira A. Fulton Schools of Engineering at Arizona State University, Jan 2025 - Oct 2025.
  • M.S. Information Technology, Arizona State University, Aug 2024 - May 2026.
  • B.Tech Computer Science and Engineering, M. S. Ramaiah University of Applied Sciences, Aug 2019 - Aug 2023.

Publication

A Personalized E-Learning System Using Reinforcement Learning Through Satellite
IEEE INDICON 2023 / IEEE Xplore document 10440852
https://ieeexplore.ieee.org/document/10440852

Certifications And Recognition

  • AWS Academy Graduate - Cloud Security Foundations training badge.
  • Technology Innovation Lab - Honeywell Aerospace & ASU.
  • KSCST 46th Series Student Project Programme Research Grant.

Open To Roles

I am currently open to full-time opportunities in AI Security, AI/ML Security, Product Security for AI, Security Research Engineering, and AI Systems Security, particularly roles involving agentic AI, LLM applications, MCP/tool security, identity and access control, model supply-chain security, adversarial testing, and secure AI infrastructure.

Location: Greater Phoenix Area, Arizona, USA
Availability: Available now; open to relocation
Work authorization: F-1 OPT; H-1B sponsorship required for longer-term employment
Contact: [email protected]


Security claims here are limited to public repositories, committed artifacts, documented limitations, and reproducible checks. Research prototypes are not described as production deployments.

Pinned Loading

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

    Monitor MCP tool calls for prompt injection, PII leakage, shadow servers, and exfiltration patterns.

    Python 14 4

  2. hf-model-provenance-scanner hf-model-provenance-scanner Public

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

    Python

  3. llm-redteam-framework llm-redteam-framework Public

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

    Python

  4. 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