Engineering leaders from Martin Marietta Materials, ChargePoint, Amazon and Analog Devices assess self-hosted judging platforms built in 72 hours, required to run offline and to keep their security promises under live probing
Hackathon Raptors, a UK Community Interest Company (CIC #15557917), has published the results of DOGFOOD 2026, an online hackathon held from September 26 to September 29, 2026, with an unusual premise: every team built the same product, a self-hosted platform for running a hackathon, and the winning platform is meant to be forked and used for a real event. Teams were building the system that could one day judge them. 194 projects were ranked in the final standings.
The brief set a high bar for something built in three days. A single docker compose up had to bring up a working portal on a laptop, with no cloud account and no hosted service, and the system had to keep working with the network switched off once built. Role isolation had to be enforced on the server rather than in the interface, and any adjustment of judges’ scores had to be documented and defensible. Features were arranged in a four-tier ladder, from accounts, teams and deadline-enforced submissions, through judge assignment, weighted rubrics and score normalization, to community voting with abuse controls and API-first stretch goals.
Evaluation Framework
Submissions were scored on four weighted criteria: Tier Completion and Correctness at 40 percent, Judging Integrity at 25 percent, Adoptability and Operability at 20 percent, and Code Quality and Innovation at 15 percent. The brief rewarded honest gap reporting, so a team that claimed a tier it had not built stood to lose more than a team that stated plainly where it stopped.
The final ranking treated every judge as one equal vote. A judge who scored consistently high or low was adjusted for, and a project reviewed by fewer judges was pulled slightly toward the field average. The published results show each project’s rank range under resampling rather than a single number, and every score from every judge is public on the results page.
Expert Panel
The panel was drawn from engineers who build and operate enterprise systems, and its members read submissions through the failures they see in their own work.
Hari Krishna Bethanaboina, a Lead Solutions Architect at Martin Marietta Materials, leads procurement and capital asset maintenance programs governing more than $8 billion in assets across more than 400 sites. Earlier in his career he designed a hybrid online and offline ERP model that kept 38 offshore drilling rigs in 13 countries running without continuous connectivity. He is an IEEE Senior Member and an Oracle ACE Associate. He did not only read the code in his batch. He built and ran submissions, ran the official checker, and probed access controls from the server side. Of one submission he wrote that “authentication is effectively absent: I found five separate ways to become any user, including organizer and platform owner, which makes its score-privacy claim hollow.” Of the strongest in his batch he wrote: “Its claims are also the most honest.”
Narasimha Reddy Annapareddy, Manager of NetSuite Business Systems at ChargePoint, has more than 14 years in enterprise software, specializing in integrations across finance and supply chain operations through NetSuite, Salesforce, Stripe, Workday and EDI. He is an IEEE Senior Member and a member of the IET. His lens was the data layer: whether records survive export and import, and whether an organizer could move an event in and out of the platform without loss, which the brief scored under Adoptability and Operability.
Rushikesh Hayatnagarkar, a Senior Product Manager at Amazon and a hands-on builder of technical infrastructure and autonomous multi-agent systems, brought an operator’s question to the brief: could an organizer run this platform on a Monday morning without calling the team that built it. He holds an MBA from Arizona State University and an MS in Analytics from Georgia Tech, and mentors founders through Alchemist Accelerator and ADPList.
Rupesh Kumar Prasad, a Technical Architect and Senior Manager at Analog Devices, has led a global team building large-scale, performance-critical applications for more than ten years. His background spans identity and single sign-on across OAuth, OpenID Connect and SAML, REST API architecture, cloud migration and web application security. He is an IEEE Senior Member and a published author on browser privacy and authentication. His expertise maps onto the brief’s foundation: a five-role model, sessions and authentication that every other guarantee depends on.
Top Submissions
First place went to Verdict, with a final score of 4.611 out of 5.00. Verdict is a self-hostable platform built around what its team calls judging you can defend: one command brings up registration, teams, submissions, judging, community voting, results and certificates, fully offline. It corrects for harsh and generous judges with an additive leniency model and reports a rank range for every project instead of false precision. Verdict also won the Best Judging Engine award, chosen after the organizers ran six finalists live against four checks: assignment, normalization, role isolation and audit trail. It passed all four, with a hash-chained audit trail and a normalization proof that reproduces exactly.
RAPTOR DESK, from team GSK VERIFIED, placed second at 4.583. It takes a ranking from uncertainty to a signed decision anyone can check, correcting each judge’s leniency with a bias model and estimating each project’s chance of finishing in each prize place. The team’s write-up about the build was also among the four best articles in the event’s write-up side quest, ranked blind by four AI models.
BroadSheet, from team Binary Builders, finished third at 4.428. It covers the full event with a single Docker command, including server-enforced deadlines, duplicate detection, judge isolation, cross-judge normalization, sealed community voting and an API.
ballotbench, from team Cause Of Death, finished fourth at 4.418. Its team describes isolation that is enforced in the database queries and tested with 94 scripted attacks, and a deadline that the page, the API and a database trigger all refuse to bend. Quorum, from team Quorum Sensing, finished fifth at 4.402, with a design that measures each judge’s leniency, explains every score and declines to print a podium the data cannot support.
What Separated the Field
The pattern across the panel’s notes was consistent. The platforms that scored highest were the ones whose claims matched their behaviour when tested. Lower-scoring entries rarely lacked ambition. More often they described security, normalization or tiers that the running system did not deliver, or they worked only on the machine that built them. Bethanaboina’s notes on his own batch describe submissions whose role checks held while their login could be bypassed, and documentation that, in his words, “overstates what works.”
The brief made those gaps easy to find. The offline rule meant a platform had to start from a fresh clone with the documented commands. The integrity criteria meant a judge could send requests the interface would never send and see whether the server refused them. A submission could not pass on appearance alone.
Why the Format Matters
A hackathon judging platform is a small system with a large job. Participants need to trust that their work was received on time and scored fairly. Judges need to trust that their scores are private and recorded as entered. Organizers need to be able to defend a result if anyone challenges it. Every one of those depends on identity, access control and an honest account of what the system does, which are the same foundations enterprise software depends on at far greater scale.
By asking teams to build a platform that would judge their own field, and putting it in front of engineers who run supply chains, finance integrations, freight operations and identity systems for a living, DOGFOOD 2026 made those foundations visible in 72 hours. The teams that finished on top were the ones whose work held up when someone tried to break it.
About Hackathon Raptors
Hackathon Raptors, a UK Community Interest Company (CIC #15557917), is London-based and runs technology challenges that test engineering fundamentals under real constraints. Its community brings together engineers from Amazon, Microsoft, Walmart, AWS and other major technology organizations. More information at raptors.dev or [email protected].
Feature image by Matheus Bertelli on Pexels
