AI-Assisted Software Systems

Local-intelligence-based research software, deterministic tools, infrastructure agents, conversational interfaces, model control, and deployment automation.

Focus

My AI-assisted software systems work is about using LLMs and agents to make research operations more reliable, auditable, and maintainable. I am especially interested in local-intelligence-based systems where model behavior, infrastructure access, and workflow automation can be controlled rather than treated as opaque magic.

Core Themes

• LLM-powered multi-agent lab operations.
• Control-plane and server-agent architectures for distributed infrastructure.
• Natural-language interfaces for read-only or carefully controlled system workflows.
• Prompt optimization and iterative evaluation of model behavior.
• Ranking, retrieval, personalization, and output prioritization.
• Local model serving, model switching, and deployment automation.

Representative Work

• Agentic AI Lab Manager for Research Computing: open-source-model-powered lab operations system.
• Reusable LLM pipelines and orchestration systems for modular workflows.
• Infrastructure-aware conversational interfaces with deterministic tool boundaries.
• Model behavior improvement through iterative feedback and evaluation.

Decision-Science Connection

This agenda connects software engineering to statistical decision systems. Agents are useful only when their permissions, outputs, failure modes, and review paths are explicit enough for a human team to trust the workflow.