justingrammens

justingrammens / OSN-2026

Public

Spec First, Agents Second: Engineering AI Systems with Discipline. Slides and code from the Open Source North 2026 Conference.

19
3
94% credibility
Found May 28, 2026 at 20 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

MDAM (Medication Dose Alert Monitor) is an open-source demonstration project from Open Source North 2026 that shows how to build healthcare safety software using a 'spec-first' development approach. The system monitors scheduled medication doses and alerts care teams when doses go unacknowledged past a configurable threshold, escalating to charge nurses if still not addressed. It was specifically designed to demonstrate disciplined AI-assisted engineering — using AI to build faster, but within explicit constraints that preserve architectural intent. The project includes comprehensive documentation, a complete test suite, and is structured as a teaching tool for teams wanting to adopt more rigorous AI workflows.

How It Works

1
🎤 Attend a conference talk

A developer hears Justin Grammens present on responsible AI-assisted engineering at Open Source North 2026

2
💡 See the problem the system solves

They learn about a near-miss: a patient whose insulin dose went untracked at 2:47 AM because no requirement captured the timing constraint

3
🔍 Explore the live demo system

They discover MDAM, a medication alert monitor built during the talk, and see how it makes that failure structurally impossible

4
Choose your learning path
📋
Read the requirements

See the 12 clear rules the system must follow, from detecting overdue doses to requiring authenticated acknowledgment

🧪
Run the test suite

Watch 60 tests pass, seeing exactly how the system responds at boundary moments like 14:59.999 versus 15:00.000

5
📐 Understand the architecture approach

They see how pure functions return instructions, not actions — keeping the logic clean and predictable

6
🔗 Trace every requirement to code

They notice every function is annotated with which requirement it satisfies, so nothing gets lost

🏆 Grasp spec-first engineering

They leave understanding how starting with clear, testable requirements produces software that stays maintainable and safe

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 20 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is OSN-2026?

This is a conference repository containing slides and code from Justin Grammens' talk at Open Source North 2026. The project demonstrates a spec-driven workflow for building AI-assisted systems, using a healthcare alert monitoring system (MDAM) as the live demo. It shows how to use SpecKit commands with Claude Code to build reliable software by starting with specifications instead of generating code and hoping for the best. The demo system monitors scheduled medication doses and escalates alerts through a care team pipeline when doses go unacknowledged.

Why is it gaining traction?

The hook is straightforward: AI makes bad engineering faster, not better. This repo gives you a concrete, repeatable workflow for constraining AI agents with explicit requirements. The demo is compelling because it shows a real failure mode (silent alert failures in medical systems) and proves how specification-first development prevents it structurally. The pure function approach with injected time means tests are deterministic and verify behavior without relying on system clocks.

Who should use this?

Backend and full-stack developers building systems where correctness and auditability matter. Teams integrating AI coding assistants who want guardrails instead of chaos. Anyone working on healthcare, financial, or compliance-adjacent software where "tests pass but the system is wrong" is unacceptable. It's particularly useful for engineering leads trying to introduce structure into AI-assisted workflows.

Verdict?

The credibility score of 0.949999988079071% reflects a new repository with modest visibility (19 stars), but the engineering is mature. Sixty passing tests with a 90% coverage gate, discriminated union types that make illegal states unrepresentable, and a full traceability matrix from requirements to test blocks. This is not a throwaway demo. If you're serious about AI-assisted development that doesn't trade discipline for speed, this is worth the read. The slides alone are worth your time.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.