Jason Allum

Thirty years shipping software — through DOS, Windows, Unix, web, mobile, Cloud, ML, and now AI — long enough to know the fundamentals move slower than the stack. Principal engineer on hard problems, built and ran engineering organizations up to ~35 engineers, two companies acquired. Currently VP of Engineering at a marketplace that grew from 20k to 4M+ users on a platform I rebuilt underneath it.

The work rarely stays in one layer: product questions become data-model questions, storage questions, latency questions, and eventually organizational questions. Happiest rescuing systems under real load and making them boring again — an evergreen skillset that pairs well with safe and sane use of AI.

VP of Engineering at CollX (2021-01 – Present)

Inherited a Python/Flask REST API running on a single machine and falling over at ~20k users. Rebuilt the platform underneath it while the product kept shipping.

• Platform now serves 4M+ users at 500+ requests/sec peak, clearing ten thousand marketplace orders a day.
• Migrated Python/Flask to Elixir incrementally — strangler-fig pattern with both backends sharing one MySQL database through the transition, rather than a rewrite-and-cutover. No customer-facing freeze.
• Designed GraphQL surfaces (Absinthe) replacing the REST API, and introduced service boundaries, continuous deployment, and elastic clusters on GKE.
• Rewrote the React Native client onto React Relay/GraphQL over time, then trained the mobile team to maintain and extend it independently.
• Executed structural data migrations where the obvious approach stops working once there are billions of records to move.
• Small team — still in the code on the hard problems, still mentoring the engineers around me.

VP of Embedded Systems at Partium.io (formerly Slyce) (2019-01 – 2020-01)

Stayed on after Slyce / Partium acquired Industrial Learning Machines, to guide their broader parts-recognition efforts across both hardware offerings and mobile devices.

• Expanded and adapted the systems originally built for Home Depot into offerings for European train manufacturers and other vendors of large, complicated machinery.
• Target use case: field repair technicians identifying failed or broken parts that needed replacing.

Founder / Principal Engineer at Industrial Learning Machine Co. (2018-04 – 2019-01)

Spun out of Dmgctrl client work the acquirer declined — took the Home Depot contract, built the company around it, and carried it to a second exit. SKU-level part-recognition machines using machine learning, computer vision, and photogrammetry.

• Identified parts on sight in real time on embedded NVIDIA Jetson hardware, at sub-millimeter differences across ~100k SKUs.
• Fasteners were the case that made it hard: thousands look nearly identical and the camera angles aren't always good.
• Economics worked because a new SKU could be added with one scan — that pass captured the images, generated the training data, and created the catalog entry, with no separate labeling workflow.
• Device: Swift on Linux, OpenCV, SQLite. Fleet management backend: Elixir and Postgres.

Founder / CTO at Tonic Design Co. (2014-03 – 2018-05)

Dmgctrl merged with Tonic Design, a web/mobile design shop. Continued building and leading a ~35-engineer consulting organization specializing in high-risk software rescues and transformations — consumer electronics, pharma, footwear, retail, banking. I enjoyed the rescues more than the greenfield work.

• For one bank, the customer-facing mobile app was crashing 30,000 times a day across roughly 500,000 users, and they didn't know. We added monitoring, then worked the system down — simplifying the codebase from 100k+ lines, including three different JSON libraries, to a tighter 30k, and reducing crashes to a handful per day. You never quite hit zero; there's always a device or two that got left in the rain and hasn't worked right since.
• Hired, grew, and retained the engineering organization across twelve years and an acquisition.

Founder / President at Dmgctrl, Ltd. (2006-06 – 2014-03)

Built an "A-team" of ~12 engineers as a rapid-response crew for broken and off-the-rails projects. Software initiatives failed for all the usual reasons — technical choices, management, staff departures — and we handled those problems discreetly. Every project was different.

• Took over responsibility for projects to put them back on the rails: establishing proper revision control, CI/CD, and testing practices.
• As systems stabilized, coached in-house staff on best practices, then supplied ongoing support and advice to prevent recurrences.

Senior Software Engineer at Ticketleap.com (2005-11 – 2006-04)

Short-term contract to improve performance of the ticket reservation and assignment system, which was suffering under load due to design choices. Identified and corrected.

Founder / Principal Engineer at Ripit (2004-04 – 2006-06)

Built and shipped a DVD ripper for the Mac, designed around a custom, reverse-engineered DVD virtual machine. Paired with a disc-fingerprinting system I designed so recipes for rips could be saved as discs were successfully backed up. Users reported failing discs, which I would purchase to improve the virtual machine implementation — usually unlocking other problematic discs in the process. Open-sourced as dvdkit and udfkit.

Senior Software Engineer at Microsoft / Avenue A / Razorfish (2003-01 – 2005-12)

Contractor working across a large portfolio of agency clients in a number of technologies.

Software Engineer at Various (1996-01 – 2003-12)

Software engineering roles across desktop, embedded, and web.

OJT