Sadia Afroz

Leading development of agentic AI systems for financial decision-making, including tool-use orchestration, long-context reasoning, and real-time inference pipelines. Designed and deployed evaluation frameworks for high-stakes AI systems and built governance systems for safe deployment of autonomous AI agents.

Architected and deployed a large-scale AI assistant platform serving 500M+ users, including model orchestration, retrieval, and tool integration. Built LLM-as-a-judge evaluation system, reducing manual labeling by 90% and enabling rapid iteration across model variants.

Secured $6M+ in funding (DARPA, NSF) for adversarial ML, robustness, and security-critical systems. Published 45+ papers (4,000+ citations) in top-tier venues and led teams translating research into deployable ML systems.

Built ML systems for large-scale classification and anomaly detection (malware, misinformation), optimizing for precision/recall tradeoffs in adversarial settings. Led research on adversarial ML and robustness, directly influencing production model hardening.

Pioneered work in adversarial ML and authorship attribution. Developed early techniques in privacy-preserving machine learning and stylometry.

Built NLP-based clustering systems for large-scale app similarity detection.

Developed secure VoIP systems with encrypted communication.

Ph.D. in Computer Science from Drexel University, completed in 2014.

B.Sc. in Computer Science and Engineering from Bangladesh University of Engineering and Technology, completed in 2007.