How can we systematically detect, quantify, and mitigate the reliability and behavior variability challenges in software systems that integrate LLMs? This research would explore novel testing methodologies for LLM-dependent software, focusing on developing deterministic test frameworks for fundamentally non-deterministic components, creating automated regression testing strategies, and establishing industry standards for LLM API robustness evaluation.
Extended the StegFormer architecture to hide secret images across multiple cover images, with innovative channel- and spatial-splitting strategies, parallel processing branches, and a composite loss function for robust training.
Collaboration: Working with Professor A. B. M. Alim Al Islam from BUET, whose research covers wireless networking, embedded systems, HCI, security and privacy.
Read MoreDeveloped an extended taxonomy for API contracts in LLM libraries, conducted empirical studies on real-world issues, and proposed recommendations for LLM library developers and LLMOps tool builders.
Collaboration: Collaborating with Dr. Akond Rahman, Assistant Professor at Auburn University, who specializes in DevOps, cybersecurity, and secure software development.
Engineered an advanced blockchain framework with sharding, Layer-2 solutions, and a DAG ledger, implementing patient-centric consent management, HL7 FHIR standards, and privacy-preserving computation.
Collaboration: Project supervised by Professor ASM Latiful Hoque from BUET (my undergraduate thesis supervisor), who specializes in data warehousing, data mining, big data analytics, and database technologies.
Read MoreDeveloped Bangladesh's first anonymous online reporting system and analyzed crowd-sourced reports using transformer models and NRC Lexicon-based analysis.