Research to Practice

We publish at top AI conferences.
Then we apply it to real infrastructure.

Why Research Matters

Most infrastructure companies rely on best practices and experience. We add mathematical rigor and AI innovation.

Our research isn't academic exercise - it's the foundation for better infrastructure delivery.

ICLRNeurIPSACLAAAI

Publications

2024-2025

CostNav: A Navigation Benchmark for Cost-Aware Evaluation of Embodied Agents

Haebin Seong*, et al. (*equal contribution)

arXiv, November 2024

Economic modeling for autonomous systems. First benchmark evaluating robots by profit, not just task success.

Relevance to Infrastructure: Cost modeling methodology applicable to datacenter and infrastructure optimization.

HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models

Seanie Lee*, Haebin Seong*, et al.

ICLR 2025

AI safety through efficient knowledge distillation. Smaller, faster safety models without sacrificing accuracy.

Relevance to Infrastructure: Efficient AI deployment for resource-constrained environments.

SafeRoute: Adaptive Model Selection for Efficient and Accurate Safety Guardrails

Seanie Lee*, Dong Bok Lee*, et al. (Haebin Seong co-author)

ACL 2025

Dynamic model selection for AI safety applications.

Relevance to Infrastructure: Adaptive resource allocation for optimal performance-cost trade-offs.

D2E: Scaling Vision-Action Pretraining on Desktop Data for Transfer to Embodied AI

Suhwan Choi*, Jaeyoon Jung*, Haebin Seong*, et al.

arXiv, 2024

Transfer learning from desktop environments to robotics. 96.6% success on manipulation tasks.

Relevance to Infrastructure: Transfer learning approaches for infrastructure automation.

Research Areas

🤖 Embodied AI & Robotics

  • • Economic modeling
  • • Cost-aware decision making
  • • Navigation and automation
  • • Vision-action learning

🛡️ AI Safety & Security

  • • LLM guardrails
  • • Red-teaming and jailbreak defense
  • • Efficient safety models
  • • Secure AI deployment

⚙️ Infrastructure Optimization

  • • Cost modeling frameworks
  • • Resource allocation
  • • Predictive analytics
  • • Performance optimization

From Research to Product

How We Apply Research

1. Publish & Validate

Peer-reviewed research at top conferences. Academic rigor, proven approaches.

2. Prototype & Test

Apply to internal infrastructure first. Validate in controlled environments.

3. Pilot Deployment

Early customer projects. Real-world validation and refinement.

4. Production Service

Full service offering. Proven, reliable, scalable.

Current Stage:
• Cost modeling: Pilot deployments ✓
• Predictive maintenance: Early testing
• Infrastructure automation: Research → Prototype

Research Background

Our team members bring experience from these institutions:

  • • KAIST (Machine Learning and AI Lab)
  • • Maum.AI (WoRV Team)
  • • Theori (AI Security)
  • • Yonsei University
  • • Postech (Laboratory for Unix Security)
Let's Discuss Your Project