Md Ajwad Akil
Md Ajwad Akil
CS Ph.D. Student
Purdue University
ajwadakillabib@gmail.com
CV

Md Ajwad Akil

PhD student in Computer Science, Purdue University

I'm a 4th-year CS PhD student at Purdue, advised by Elisa Bertino. My research interest sits at the intersection of machine learning and security: adversarial malware and evasion of ML-based detectors, LLM and agent security (prompt injection, code-hallucination detection), and reasoning over low-level code.

I am working as a Research Engineer Intern at Samsung Research America this Fall and also worked as a Machine Learning Engineer Intern at DoorDash. Before Purdue, I earned my B.Sc. from BUET and taught as a lecturer at United International University, Dhaka, Bangladesh.

I like working on both sides of security - attack and defense. If that resonates, feel free to reach out.

News

Publications

Experience

Research Engineer InternMountain View, CA
Samsung Research America · Mobile Platforms and Solutions Lab · Aug 2026 – Present
  • Built an automated benchmarking pipeline evaluating LLM agents across on-device mobile tasks with real app integrations.
  • Ran indirect prompt injection red-teaming of on-device agents: designed cross-app attack scenarios and executed end-to-end attacks to probe robustness.
Machine Learning Engineer InternSan Francisco, CA
DoorDash · New Verticals, Agentic Catalog · May 2026 – Aug 2026
  • Integrated an in-house MCP web-search service into a catalog-enrichment agent with observable multi-round tool calling.
  • Fine-tuned Qwen3.5 9B/27B VLMs (LoRA, assistant-only loss masking, BF16, multi-GPU DDP) for grounded tool use and attribute extraction.
  • Reached 95.5% precision; the 27B model beat Gemini's attribute coverage by 1.18pp while cutting serving-and-search cost by 43%.
Graduate Research AssistantWest Lafayette, IN
Cyber2SLab, Department of Computer Science, Purdue University · Aug 2023 – Present
  • PhantomCall — black-box FCG-perturbation attack on Windows PE malware; 85–100% attack success across three detectors.
  • LLMalMorph — LLM-driven functional malware variant generation; Best Paper, IEEE TPS 2025.
  • LLM code security — training-free method contrasting an auditor LLM's security- vs. style-review signals to select more secure LLM-generated code and localize likely-vulnerable lines, outperforming a frontier LLM judge at selection.
AI Engineering InternDhaka, Bangladesh
Intelligent Machines · Jul 2021 – Jun 2022
  • Built noise-robust audio event detection and multilingual OCR pipelines (≤12% WER), deployed on Azure as a POC.

CV

Full education, teaching, awards, and skills below. For the formatted one-page version:

Download PDF

Education

Purdue UniversityIndiana, USA
PhD in Computer Science (CGPA 3.90) · MS in Computer Science (CGPA 3.84) · Advisor: Dr. Elisa Bertino · Aug 2023 – Present
Bangladesh University of Engineering and TechnologyDhaka, Bangladesh
B.Sc. in Computer Science and Engineering · 2017 – 2022

Teaching

Graduate Teaching AssistantPurdue University
CS 18000: Problem Solving and Object-Oriented Programming · Fall 2023 – Spring 2026
LecturerUnited International University
Computer Architecture · DSA II · OOP · Structured Programming · Jul 2022 – Jul 2023

Service & Awards

AwardsBest Paper Award, IEEE TPS 2025 · Distinguished Artifact Reviewer, ACSAC 2024 · Graduate Teaching Award, Purdue CS 2024
Sub-reviewerNDSS 2025 · ACSAC 2024 · RAID 2024 · IEEE/ACM Transactions on Networking
CommitteeArtifact Evaluation Committee, ACSAC 2024

Technical Skills

LanguagesPython, C++, Java, JavaScript
ML / LLMPyTorch, Hugging Face, TRL, vLLM, SGLang, LangChain, LangGraph, smolagents, scikit-learn
Infra / DataAWS, SageMaker, Modal, Docker, Git, PostgreSQL, MongoDB, Elasticsearch, ChromaDB
SecurityIDA Pro, Binary Ninja, LIEF, pefile, CodeQL, FlowDroid, MobSF
Email copied