AI Security Researcher based in Abuja, Nigeria. I study how retrieval-augmented generation (RAG) systems and LLMs can be attacked, and how to defend them with methods that are light enough for real-world, low-resource settings.
I also founded Khemshield, a cybersecurity and AI training organisation that has reached over 264 registrations across eleven Nigerian states.
- RAG & LLM Security: corpus poisoning, and why signed sources can still be stale or untrustworthy in RAG-based cybersecurity assistants
- AI Security & Trustworthiness: building AI systems people can rely on, and testing where that trust breaks
- Adversarial Machine Learning: how attackers manipulate the data and inputs that ML systems rely on
- Privacy-Preserving Machine Learning: homomorphic encryption, federated learning, and differential privacy for sensitive data
- BSc Cybersecurity, Bayero University Kano. Best graduating student, CGPA 4.94/5.00
- Published: Privacy-Preserving Data Analysis Using Homomorphic Encryption, Bayero Journal of Physics and Mathematical Sciences
- In progress: privacy-preserving sickle cell baseline prediction using HE, federated learning, and differential privacy
- In progress: phishing detection and user education study (abstract submitted to a Wiley publication)
| Project | Description |
|---|---|
| rag-corpus-poisoning-iot-security | Corpus poisoning attacks on RAG-based IoT and threat-intelligence assistants, with a lightweight provenance defence |
| privacy-preserving-heart-diagnosis | Heart disease prediction on Paillier-encrypted data, so the server never sees patient data or keys |
| rag-bootcamp-projects | Hands-on RAG engineering: hybrid search, agentic pipelines, GraphRAG, and evaluation |
I keep my study work in the open:
- gen-ai-training: applied NLP, deep learning, and agentic systems
- security-training: cryptography and core security concepts
I'm open to collaborating on RAG and LLM security, adversarial ML, and applied security research. Email: abdulkareem@khemshield.com