The Adimension

Bridging human ingenuity and machine intelligence

Overview

The Adimension develops transparent, inspectable, and human-centered systems at the intersection of clinical medicine, scientific data, and machine intelligence.

All work is grounded in the DEITY Principles Framework (Data • Ethics • Informatics • Technology • You), published at:

DEITY Principles

Principle	Operational Definition
Data	Transparent, multi-source, reusable, and anonymized inputs with clear provenance
Ethics	Continuous evolution of ground truth that respects operator variability and clinical accountability
Informatics	Strictly structured, interoperable outputs (JSON polygons, executable traces, metrics)
Technology	Resource-conscious methods (4-bit quantization, single-GPU loops, infrastructure-agnostic design)
You	The human remains the active collaborator; the system never issues a final clinical statement

Active Repositories

ATRIA-EchoTrace

Fine-tuning of Google MedGemma-1.5-4B-it (QLoRA) to output clinically usable 30-point polygons for left-ventricle and left-atrium endocardial borders. Full pipeline: DICOM/AVI → PNG + JSON → QLoRA → inference → interactive four-panel human-in-the-loop revision
Structured JSON coordinates + publication-ready overlays
Model proposal is never the final answer; the clinician retains full edit control

Agent-ProSAT

Agent-driven loop (inspired by Karpathy’s autoresearch) that iteratively improves pure-Python solvers for the NVIDIA Nemotron Reasoning Challenge (Alice-in-Wonderland puzzles). Agent never generates free-form CoT
Twin-blind verification (solver prohibited from seeing the ground-truth answer)
Produces short, fully re-executable reasoning traces
Cited twice in the official NVIDIA Technical Blog for emphasizing verified traces over unfiltered scale

REDACTS

Contract-driven static and dynamic forensic toolkit for REDCap deployments. Baseline-driven differential analysis to detect tampering, backdoors, and patterns such as INFINITERED. Explicitly not a replacement for manual review.

Codex Editoris VitAI:

Transforms scientific articles into interactive objects for digital networking, impact evaluation, utilising Europe PMC, local AI models for private, transparent post-publication and publication lifecycle evaluation.

Design Stance

Every system is built so that: Intermediate steps remain inspectable and re-executable
The human can intervene at any point without breaking the pipeline
Outputs are structured for both clinical use and further machine learning (SFT, RL, active learning)
Resource constraints of real clinical and research environments are respected

Correct answers that cannot be audited are treated as incomplete.

Founder

Shehab Anwer, MD
Cardiologist / Researcher · REDCap System Administrator at Magdi Yacoub Heart Foundation · Editor at European Heart Journal – Imaging Methods and Practice, European Heart Journal – Cardiovascular Imaging, International Journal of Cardiovascular Imaging

Contact & Collaboration

Organization email: atrium@theadimension.com
Website: www.theadimension.com
All repositories are released under Apache-2.0 unless otherwise noted

Open invitations for technical collaboration, clinical validation, and joint experimentation are welcome. Preferred channels are GitHub Issues, Pull Requests, or the public discussion threads linked above.