Applied Research Institute

Applied research at the intersection of AI and the physical world.

From trustworthy AI evaluation to embodied systems that operate in clinical, industrial, and molecular environments. Independent research with open methodology.

Our Research

Independent research, compounding infrastructure

Our research spans two converging frontiers: ensuring AI systems are trustworthy and safe, and applying AI to physical systems that interact with the real world.

H.H.A. Applied Research Institute is an independent California nonprofit. From adversarial robustness testing to magnetically guided microrobots, we pursue problems where rigorous methodology and interdisciplinary expertise produce verifiable results.

Every finding is verifiable. Every tool is public.

8
Published Research Papers
100%
Open Source
501(c)(3)
Tax-Exempt Nonprofit

Principal investigators

Haedar Hadi

Haedar Hadi

Lead Principal Investigator

MS Computer Science (Boston University, Information Systems). ML model development, reinforcement learning architectures, and evaluation methodology. Leads benchmark development, multi-agent system architecture, and reproducible evaluation pipelines. In embodied AI: autonomous navigation controller development, sim-to-real transfer, and safety-constrained policy optimization.

Hass Dhia

Hass Dhia

Co-Principal Investigator

MS Biomedical Sciences, medical school background. AI infrastructure architect with production systems at scale. Leads experimental design, adversarial evaluation methodology, and analysis. In embodied AI: biomedical domain expertise in anatomy, physiology, and pharmacokinetics. Regulatory strategy and clinical problem framing.

Ahmed Dhia

Ahmed Dhia

Key Team Member

Director of Manufacturing. Dataset curation, longitudinal safety constraint analysis, and experimental coordination. In embodied AI: design for manufacturability (DFM), production scaling, and quality systems (ISO 13485). Bridges laboratory prototypes to production-ready systems.

Published research

Open-source reinforcement learning environments for safety-critical medical applications. All packages include comprehensive test suites and are available on PyPI.

1
AnestheSim
RL environments for automated anesthesia drug dosing (109 tests)
2
NeuroSim
RL platform for brain-computer interfaces (158 tests)
3
VascularSim
Microbot navigation in patient-derived vascular networks (139 tests)
4
PeptideGym
RL environments for therapeutic peptide design (125 tests)
5
OncoSim
RL for radiation therapy treatment planning (141 tests)
6
CardioSim
RL environments for cardiac electrophysiology management
7
GlucoSim
RL environments for glucose management and insulin dosing
8
VentiSim
RL environments for mechanical ventilation control
View all publications →

Research originally published by Smart Technology Investments Research Institute (smarttechinvest.com/research). Licensed to H.H.A. Applied Research Institute under a non-exclusive research license for R&D and grant pursuit. Commercial exploitation rights retained by STI.

Core research programs

Spanning AI safety evaluation, adversarial robustness, embodied systems, and governance.

Evaluation & Adversarial Testing

Existing AI benchmarks evaluate models in isolation under static conditions. We develop reproducible evaluation frameworks with granular, binary-testable criteria that capture real-world failure modes, combined with red-team methodologies and multi-perspective stress-testing protocols that systematically discover vulnerabilities before deployment.

Multi-Agent Safety & Governance

How do safety failures propagate across networks of cooperating AI agents, and how should governance frameworks respond? We study emergent risks invisible to single-agent evaluation — failure cascades, unintended coordination, and constraint erosion — while producing empirical evidence on compliance gaps and democratic oversight mechanisms.

Embodied AI Systems

Research across five scale tiers — from orbital platforms to molecular machines. We identify and evaluate opportunities where AI-driven control, manufacturing engineering, and domain expertise converge to bridge the gap between laboratory prototypes and deployable systems.

RL for Medical Decision-Making

Open-source reinforcement learning environments for safety-critical medical applications — anesthesia dosing, ventilation control, glucose management, radiation therapy, and cardiac electrophysiology. Eight published simulation platforms with over 1,150 combined tests, all available on PyPI.

How we work

Verifiable results through structured methodology.

1

Criteria-Driven Evaluation

Research questions decompose into discrete, binary-testable criteria before investigation begins.

2

Multi-Perspective Analysis

Each finding is examined through independent analytical lenses to surface blind spots.

3

Adversarial Stress Testing

Results undergo structured red-team review through automated pipelines at scale.

4

Open Reproducibility

Every benchmark, dataset, and framework ships as open-source software.

Collaborate with us

We welcome partnerships with researchers, institutions, funding agencies, and organizations working on AI safety, governance, and evaluation.