🎓 Ph.D. Candidate in Computer Science – AI/ML | 🤖 ML Intern @ HP
📍 Based in Austin, Texas | 🌐 saharzargarzadeh.github.io
I'm a Ph.D. researcher specializing in adversarial machine learning, power-based side-channel analysis, and IoT security. My work focuses on building robust anomaly detection systems using multimodal data (e.g., power traces, system calls), and enhancing them with transformer-based architectures and explainable AI.
💼 I'm currently interning with the Personal Devices team at HP, incorporate Microsoft Threat Modeling practices into secure and interpretable AI/ML pipelines using Vertex AI and TensorFlow.
If you're interested in my research work on AI robustness and adversarial attacks on IoT devices, check out my latest post:
🔐 Securing AI/ML Models in IoT Devices: Lessons from Power-Based Adversarial Attacks
- Languages: Python, C#, Java, SQL, JavaScript, jQuery
- AI/ML Tools: TensorFlow, PyTorch, BiLSTM, Attention-based CNN , SHAP, PCA, GANs, Transformers, Huggingface
- Systems & Platforms: FreeRTOS, Google Colab, .NET Core, Docker, Git
- Cloud & Deployment: Google Cloud (Vertex AI), REST APIs, MongoDB, MySQL
- Special Topics: GenAI, Side-Channel Attacks, Adversarial Training, IoT Malware, threat modeling, Power Signature Analysis
-
🧠 Featured Project: AWS Agentic AI Workshop
🚀 aws-agentic-ai-workshop
_Build cost-efficient, structured LLM agents using SageMaker, Bedrock, CrewAI, and Langfuse — complete with visualizations, tool chaining, and observability. -
🛡 PowerAdversarial-ML
Adversarial dummy code injection to evade power-based AI malware detectors.
Implements LSTM, BiLSTM, CNN+Attention, and Autoencoder+MLP architectures with SHAP-based feature targeting. -
🛡 Explainable AI Assistant (GenAI 2025)
🤖 Explainable Gemini-powered assistant for IoT power anomaly detection
Built with Retrieval-Augmented Generation (RAG), embeddings, and few-shot prompting as part of Google’s 5-Day Generative AI Capstone.
-
🔍 Semantic Search for Movies (Hackathon6)
🚀 semantic-movie-search
Built a Streamlit app using Huggingface + Pinecone for semantic movie recommendations. -
🧠 Image Captioning with CNN + Transformer
Used EfficientNet with LSTM Transformer Decoder to generate visual captions with high BLEU scores. -
📋 LLM Profile Matcher
🚀 llm-profile-matcher
AI agent for expert-matching competition using embeddings + reasoning from vector similarity.
- 📡 Beam Profiling and Beamforming Modeling for mmWave NextG Networks, ICCCN 2023
- 🚗 Cooperative Spectrum Sensing for CR-VANET, Elsevier Vehicular Communications 2020
- 🤝 Cross-Cultural e-Negotiation System, ICTOK 2016
- 🎤 Speaker at SCRF 2024 & UTA CSE 50th Anniversary Ph.D. Talks
- 📜 Google Cloud Skill Badge: Build & Deploy ML on Vertex AI (2025)
- 🐳 Docker Foundation Certificate (2024)