deako

How to write safety critical code

Tue, 05 May 2026 00:00:00 +0000

Software fails in ways that other engineered artifacts simply do not. When a bridge collapses under a heavy load, it fails visibly, leaving behind physical evidence that a post-hoc analysis can usually reconstruct. Software, by contrast, fails silently and conditionally. And some bugs emerge only under a precise, malicious combination of inputs that occurs once in every ten million executions.

Some NASA examples:

The Mars Climate Orbiter (1999): Lost because two subsystems exchanged data using different unit conventions.

On staying a beginner

Sun, 15 Mar 2026 00:00:00 +0000

Being a physician-researcher means being a beginner in two places at once, more or less indefinitely.

In the clinic, I know enough to be useful and not enough to stop learning. In the lab, the same. The overlap between the two (the territory where I’m supposed to have something to contribute) is a narrow strip that I’m still figuring out the shape of.

This used to feel like a problem. It now feels like the point.

Designing for uncertainty

Wed, 18 Feb 2026 00:00:00 +0000

A common framing in clinical decision support is the information problem: the physician doesn’t have what they need, so the system should surface it. If we can just get the right data in front of the right person at the right time, better decisions follow.

This framing is wrong, or at least incomplete.

Most decisions aren’t data-limited

In my experience at the bedside, the hard calls aren’t usually the ones where I lack information. They’re the ones where I have a lot of information that points in different directions, and I need to act anyway, often quickly, often irreversibly.

Wearable Sensing for Remote Health Monitoring

Sun, 01 Feb 2026 00:00:00 +0000

My PhD research at Georgia Tech focuses on wearable and ubiquitous sensing systems for remote health monitoring and diagnostics, with a particular emphasis on contexts where conventional medical infrastructure is absent or unreliable.

The core problem

Most medical sensing technology is designed around access: access to power grids, to trained technicians, to hospital-grade calibration. That design assumption quietly excludes the populations who need monitoring most. Remote and underserved communities face a compounding disadvantage: higher burden of chronic disease, less access to routine care, and diagnostic tools that weren’t built for their environment.

How I take notes

Sat, 10 Jan 2026 00:00:00 +0000

I’ve been using Obsidian as my primary note-taking environment for a few years now. This is an attempt to document what I actually do, partly for anyone curious, partly to force myself to be honest about what’s working.

The core idea

Every note is either a reference (something external I’m capturing) or a thought (something I’m developing). References link to thoughts. Thoughts link to each other. Over time the graph either reveals something or it doesn’t; both outcomes are informative.

Tangible Neuroanatomy

Sat, 01 Mar 2025 00:00:00 +0000

Neuroanatomy is notoriously hard to learn from two-dimensional diagrams. The brain is a three-dimensional structure with spatial relationships that only become intuitive through manipulation. This project explores what it looks like to build learning tools that exploit that physicality.

Approach

The project combines 3D-printed brain region models with an AR overlay that surfaces labels, functions, and clinical correlates when a learner holds or examines a physical model. The physical act of rotating a structure, finding where two regions abut, and tracing a pathway through your hands encodes spatial information differently than any screen can.

3D Rendering Engine

Fri, 01 Nov 2024 00:00:00 +0000

Bedbug is a real-time visualization engine built from scratch in C99 and Vulkan, targeting anatomical model rendering and medical imaging data for educational and diagnostic applications.

Why build this

Medical education and diagnostics both depend on spatial understanding of three-dimensional anatomy. Most existing tools either lock you into expensive proprietary software or sacrifice real-time interactivity for data fidelity. Building a renderer from scratch (with a graphics pipeline I own entirely) means I can optimize specifically for this domain: variable-resolution anatomical meshes, multi-planar DICOM reconstruction, and overlays that make sense for clinical reading rather than gaming.

Medical Education Platform

Thu, 01 Sep 2022 00:00:00 +0000

MedStudForty is a full-stack platform built to improve preclinical diagnostic reasoning through USMLE and COMLEX examination-style practice. Built from 2022 through 2026 alongside active tutoring work, it was iterated against real student performance data.

What it does

Timed clinical question blocks emulating USMLE and COMLEX examination formats
Real-time testing interface with integrated user authentication and session analytics
Custom content management system for question authoring and editing
Instructor dashboard for tracking individual student performance and identifying weak areas
SEO-optimized delivery so students can find topic-specific resources organically

What drove the design

Existing board prep tools (UWorld, Amboss) are closed systems. Building the platform myself meant I could instrument it exactly, tracking not just whether a student got a question right, but how long they spent, where they hesitated, and which conceptual gaps kept reappearing. That data shaped both the tutoring and the question bank.

Lung Cancer Case Study

Thu, 01 Jul 2021 00:00:00 +0000

During my general surgery clerkship at Detroit Medical Center Sinai-Grace Hospital in 2021, I identified an unusual presentation of lung adenocarcinoma with soft tissue metastases and proposed developing it into a formal case report.

The case

Soft tissue metastasis from lung adenocarcinoma is rare enough that isolated case reports remain clinically valuable. The presentation I encountered during clerkship was atypical in its pattern of spread, which prompted the question of whether it warranted formal documentation and publication.

Spinal Cord Injury Research

Tue, 01 May 2018 00:00:00 +0000

From 2015 to 2018, I worked across three labs at Central Michigan University studying therapeutic strategies for spinal cord injury, with a focus on mesenchymal and neural stem cell transplantation. The work progressed from entry-level research assistance to an independent honors capstone with two peer-reviewed publications.

What we were trying to do

Spinal cord injury results in permanent motor and sensory deficits because the adult central nervous system has minimal capacity for self-repair. The labs I worked in were investigating whether transplanted stem cells could create a more permissive environment for recovery, by secreting growth factors, myelinating damaged axons, or bridging the injury site.

Bio

Mon, 01 Jan 0001 00:00:00 +0000

CV · GitHub · LinkedIn

Zachary Deak, DO

Welcome to my digital garden. I’m Zach, though you might know me as Deako.

I’m a medical school graduate and self-taught programmer. I’m drawn to problems that sit at the crossroads of clinical need, hardware and software limits, and human context. I don’t fit neatly into either world anymore, and I’ve stopped trying.

My interests live at the intersection of wearable and ubiquitous sensing, signal processing, and human-centered design, with a particular pull toward tools that work where conventional medical infrastructure doesn’t reach. Sustainability isn’t an afterthought for me. I treat it as a core design constraint from the very first sketch.

Biomedical & Technical Consulting

Mon, 01 Jan 0001 00:00:00 +0000

← Back to Consulting

I hold a DO degree and am currently a PhD student in Human-Centered Computing at Georgia Tech. My background spans clinical medicine, biomedical science, and computing, with a focus on health technology design, point-of-care sensing, and clinical workflow research. If your project touches those areas and you need someone who can hold both the technical and clinical frame at the same time, I may be able to help.

Career & Application Advising

Mon, 01 Jan 0001 00:00:00 +0000

← Back to Consulting

I’ve navigated the path from pre-med to medical school to doctoral research. I know what committees are looking for and how to help you present your work honestly and compellingly. I work with applicants at every stage – from building a competitive profile early on to refining the final draft of a personal statement.

Disclaimer: I am not affiliated with any medical or graduate school admissions committee, and I cannot guarantee outcomes. Advice reflects my own experience and perspective.

Portfolio

Mon, 01 Jan 0001 00:00:00 +0000

A short set of things that give the clearest picture of how I think and what I build. For the full picture, see research, projects, and essays.

Wearable Sensing for Remote Health Monitoring

Research · 2026

My PhD research direction at Georgia Tech. Focuses on wearable and ubiquitous sensing systems for remote diagnostics in resource-limited environments, where conventional medical infrastructure is absent and the design constraints are real rather than hypothetical.