Skip to content
NB Tech NewsSubscribe

Startups4 min read

N.B. Indie Builds: A 523-Member Alert App, Oscillator AI and a Benchmark That Reports Its Own Miss

When police radio went encrypted, a Fredericton veteran built his own alert app. Plus oscillator-based machine learning from Moncton, Zwift club stats from Shediac, open AI research that publishes its misses, and source control for SQL Server.

Closing card: a New Brunswick outline with amber pins on Fredericton, Moncton and Shediac beside the headline N.B. indie builds, Oct. 21, 2026, edition 3, and five project names: Sonar 506, PRIN, ZLDR Event Reporter, OES-Resilience and MigrationOps

When police and fire radio went encrypted, a Fredericton veteran built his own alert app. A Moncton developer rebuilt his machine-learning framework from scratch, with the math in Rust. A Shediac runner turned his Zwift club's event data into reports.

The third batch of software New Brunswickers build on their own time also includes AI research that reports its own misses and a tool that puts SQL Server under source control. Each one can be tried in a browser or read on GitHub. Our last roundup ran Oct. 20.

Sonar 506: a scanner group's alert app

Sonar 506 showed 523 members on Oct. 21. Noah Starr, who told the Telegraph-Journal he launched it about a month ago, built it for Scanner Freddy Beach, a Facebook group the paper says has almost 50,000 members.

Starr, a Fredericton veteran who served eight years in the Canadian Armed Forces, helps run the group. With police and fire radio now encrypted, members piece together what is happening from posts. He wanted one place for it.

The free, bilingual web app installs to a phone's home screen. Members pick places anywhere in New Brunswick, get alerts for what happens there, and file their own reports. The app adds public data: NB 511 road notices, City of Fredericton closures and detours, Environment and Climate Change Canada weather alerts and river gauges, and N.B. Power outage areas, its credits page says.

That week the site showed 21 neighbour reports and 48 river gauges reporting. Starr told the paper reports have come in from as far away as Nova Scotia and P.E.I. The app warns that it is not an emergency service: call 911 for anything urgent.

PRIN: oscillators instead of attention, from Moncton

PRIN lists 3,496 Python tests and 1,577 Rust tests passing. It is Michael Maillet's from-scratch rebuild of his earlier PRINet 3.0. Maillet lists Moncton on GitHub.

PRIN, short for Phase-Resonance Interference Network, tracks several objects over time with coupled oscillators: units that fall into sync with one another. The README pitches it as an alternative to attention-based models, which it says struggle with long sequences.

The math runs in Rust, with a Python layer for PyTorch. The GPU code compiles to CUDA, Metal, Vulkan and plain CPU from one source. Damien Davison and Sacha Davison are credited as co-authors.

The project is MIT-licensed. A 1.0 release candidate went up on PyPI as prin-core on Sept. 16, and commits were still landing on Oct. 20.

ZLDR Event Reporter: Zwift club stats, from Shediac

ZLDR Event Reporter counts who registered, who showed up and who finished every event a Zwift running and riding club holds. Al Udell, who lists Shediac on GitHub and calls himself an artist, coder, runner and bassist, built it for the ZLDR club (Zwift Long Distance Runners and Riders) on the online training platform Zwift.

The app signs in with a Zwift account, pulls the club's events and tracks how each recurring series is trending. Organizers export reports as CSV, text, HTML or PDF. It runs on Android or as a web app in any browser. Version 1.12.0 came out July 1.

A companion app, ZLDR Events, lists upcoming club events and adds the ones you join to your calendar.

Udell's earlier iFit-Wolf experiments started by capturing workout data from his NordicTrack treadmill. Later versions let Zwift set the incline: the code reads Zwift's on-screen numbers with OCR and sends commands over the treadmill's Android debugging connection.

OES-Resilience: a benchmark that reports its own miss, from Fredericton

OES-Resilience tested its own anomaly detector on real NASA spacecraft telemetry, and the README says plainly that it missed its pre-stated goal. The protocol was fixed before any test data was scored. Two simpler methods scored higher, though not by a significant margin.

It is the most developed of 61 public repositories from Jean-François Brisson, founder of Spark AI NLP, who lists Fredericton on GitHub. He created 37 of them in October. His profile calls the work "disability-led applied AI research software." Most repositories are labelled research prototypes, and many run only on synthetic data.

Spark StreamGuard learned a similar lesson. After its original score lost a real-data benchmark, Brisson reworked it into a gate that validates and logs every decision a detector makes. Several repositories have archived releases with DOIs on Zenodo.

MigrationOps: source control for SQL Server

MigrationOps gives every SQL Server script a SHA-256 checksum, so a changed script stands out. Cat Fortman, a senior engineering consultant who lists New Brunswick on GitHub, builds the MIT-licensed .NET tool on her own time.

Teams keep stored procedures, views, functions and triggers as scripts in folders. Migrations run in the order of their timestamped filenames, each in one transaction. The README credits the idea to Entity Framework migrations and Redgate's SQL Source Control. The project has 46 stars and was last updated Oct. 13.

Fortman also publishes SqlServer.EngineeringToolkit, T-SQL scripts for everyday database work, from dropping a column without breaking its dependencies to diagnosing slow queries. Her GitHub profile says she spends her free time on a 2D game engine.

Know a New Brunswick side project that deserves a look? Send us a tip.

Sources

Related coverage

Tags