Mostofa Najmus Sakib

Building Intelligent AI Systems, From Research to Deployment

Generative AI · Agentic systems · Enterprise RAG · Adversarial robustness

I am an AI/ML engineer and Ph.D. candidate in Computing, with a Computer Science emphasis, at Boise State University. My focus is generative AI, tool-using agents, retrieval-augmented generation, and adversarial robustness. I work across the AI development lifecycle—from data pipelines and model experimentation to agent orchestration, APIs, evaluation, and deployment.

My recent engineering work includes an enterprise agentic RAG system for retrieving, analyzing, and comparing 15,000+ agricultural product labels. I combined LangGraph and LangChain orchestration with multi-source retrieval, LLM-based reranking, and MCP tools connecting agents to structured data and business systems. Label-comparison workflows, LangSmith evaluation and observability, and guardrails supported recommendations grounded in active ingredients, application constraints, location, and inventory.

Beyond LLM orchestration, I have developed Azure and Databricks pipelines for more than one million social-media interactions and applied sentiment and stance analysis to autonomous-vehicle public opinion. My software engineering experience includes Python APIs, cloud-hosted document automation, backend database integration, and Docker/Kubernetes deployment. I work with domain experts to turn business requirements into AI workflows and deliver iteratively through Agile development.

My PhD research at Boise State University's AI-based Security Lab focuses on human-centered AI robustness, adversarial NLP, and information credibility. I develop LLM-crafted credibility-drift attacks that manipulate linguistic, stylistic, and source-credibility cues to influence perceived news believability while preserving semantic similarity. My work includes developing LLM-based proxy models (ICL and COLA), evaluating adversarial attack effectiveness, investigating offline reinforcement learning-based defenses, and conducting human-subject studies to assess how AI-generated transformations affect credibility judgments and reasoning. I also develop graph neural network models integrating human believability signals to improve information reliability assessment. My broader research spans adversarial robustness, human-centered AI, and graph learning, with publications at WSDM, CSCW, ICWSM, and ASONAM.