Skip to content

What we do

Research Themes & Projects

Our work focuses on practical AI systems, intelligent agentic workflows, reproducible benchmarking, and human-centered software engineering.

Research Philosophy & Vision

At Aura Lab, we investigate artificial intelligence and intelligent agents with a focus on real-world utility, reproducibility, and rigorous evaluation. By bridging theoretical machine learning concepts with hands-on software engineering and systems testing, our group explores how agentic architectures can reliably assist, automate, and collaborate in complex computational domains.

Research topics

AI & Large Language Models

Reasoning, retrieval-augmented generation, evaluation, and domain-specific applications.

  1. Structured Retrieval and Factuality Verification for Scientific Domain Large Language Models

    Bert T , Ortega-Marquez J , Miljkovic N

    Aura Lab Preprints · 2025

    Tulescu Bert Tulescu Bert Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2511.00204

Intelligent Agents

Autonomous planning, tool use, structured workflows, and digital environment interaction.

  1. Aura-Bench: An Open Evaluation Suite for Multi-Step AI Agent Tool Execution

    Bert T , Miller G , Ortega-Marquez J , Miljkovic N

    Open Source Software & Datasets · 2026

    Tulescu Bert Tulescu Bert Gillian Miller, BS Gillian Miller Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2605.00331

  2. Benchmarking Agentic Tool Use and Reasoning in Repository-Level Code Comprehension

    Miller G , Bert T , Castro-Palacin A , Miljkovic N

    Aura Lab Technical Report · 2026

    Gillian Miller, BS Gillian Miller Tulescu Bert Tulescu Bert Alberto Castro-Palacin, MSc Alberto Castro-Palacin Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2602.00101

  3. Automated Regression Testing and Patch Generation with Multi-Agent Workflows: A Case Study

    Castro-Palacin A , Bert T , Ortega-Marquez J , Miljkovic N

    Aura Lab Technical Report Series · 2025

    Alberto Castro-Palacin, MSc Alberto Castro-Palacin Tulescu Bert Tulescu Bert Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access

AI for Software Engineering

AI-assisted code understanding, debugging, testing, architecture, and mobile systems.

  1. Benchmarking Agentic Tool Use and Reasoning in Repository-Level Code Comprehension

    Miller G , Bert T , Castro-Palacin A , Miljkovic N

    Aura Lab Technical Report · 2026

    Gillian Miller, BS Gillian Miller Tulescu Bert Tulescu Bert Alberto Castro-Palacin, MSc Alberto Castro-Palacin Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2602.00101

  2. Automated Regression Testing and Patch Generation with Multi-Agent Workflows: A Case Study

    Castro-Palacin A , Bert T , Ortega-Marquez J , Miljkovic N

    Aura Lab Technical Report Series · 2025

    Alberto Castro-Palacin, MSc Alberto Castro-Palacin Tulescu Bert Tulescu Bert Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access

  3. Transparency and Multi-Turn Reliability in Human–AI Software Development Workflows

    Tripathy S , Bert T , Miller G , Miljkovic N

    Aura Lab Working Paper · 2025

    Shreya Tripathy, BS Shreya Tripathy Tulescu Bert Tulescu Bert Gillian Miller, BS Gillian Miller Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access

Human–AI Interaction

Usability, reliability, transparency, and human-agent collaborative systems.

  1. Transparency and Multi-Turn Reliability in Human–AI Software Development Workflows

    Tripathy S , Bert T , Miller G , Miljkovic N

    Aura Lab Working Paper · 2025

    Shreya Tripathy, BS Shreya Tripathy Tulescu Bert Tulescu Bert Gillian Miller, BS Gillian Miller Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access

Approaches

Research Prototypes & Systems

Applied software prototypes and empirical testbeds for real-world scenarios.

  1. Aura-Bench: An Open Evaluation Suite for Multi-Step AI Agent Tool Execution

    Bert T , Miller G , Ortega-Marquez J , Miljkovic N

    Open Source Software & Datasets · 2026

    Tulescu Bert Tulescu Bert Gillian Miller, BS Gillian Miller Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2605.00331

  2. Structured Retrieval and Factuality Verification for Scientific Domain Large Language Models

    Bert T , Ortega-Marquez J , Miljkovic N

    Aura Lab Preprints · 2025

    Tulescu Bert Tulescu Bert Jorge Ortega-Marquez, MS Jorge Ortega-Marquez Nenad Miljkovic, PhD Nenad Miljkovic
    Cited by 0 Mentee-led PI first/senior Collaboration Open access arXiv:2511.00204

Open Interactive Tooling & Simulation Testbeds

Inspired by open science initiatives, Aura Lab develops accessible web tools, interactive telemetry visualizers, and simulation sandboxes for peer evaluation.

Interactive Harness

Aura-Trace Visualizer

Real-time interactive trace inspection tool for multi-step agent planning, AST traversal, and sandboxed bash/API execution logging.

v0.2.4 · Open Source View Specs →
Benchmarking Suite

RAG Factuality Checker

Automated synthetic testbed for measuring context retrieval precision, citation fidelity, and anti-hallucination verification in scientific domain queries.

v1.0.1 · Open Access View Specs →

Computing Infrastructure & Telemetry

Aura Lab operates reproducible experimental testbeds leveraging heterogeneous cloud accelerators and isolated microVM containers for safe, deterministic AI agent evaluation.

Accelerator Fleet
GPU Clusters & Cloud AI
High-throughput model serving & batch evaluation
Sandbox Isolation
Secure microVM Testbeds
Deterministic ephemeral execution & audit logging
Reproducibility
Open Datasets & Traces
Public artifacts, code repositories & telemetry