About the challenge

Getting code to run has never been easier. Getting a system to actually work, at scale, under real constraints, is still hard. That gap is what this hackathon is about.

Global Innovation Build Challenge (GIBC) V2 is a fully online, student-led international hackathon. GIBC V1 brought together 552 participants worldwide with backing from sponsors including NordVPN. V2 builds on that with three tracks, an international judging panel, and a larger sponsor lineup.

Over the 10-week build period from July 11 to October 1, 2026, you can register at any time, build solo or in a team of up to 6, and submit to one of three tracks. Projects are reviewed by a panel of industry professionals and academic experts, with winners announced at a virtual closing ceremony on October 8, 2026.

Track 01: TECH (Foundational LLM Development)

Build and train a language model capped at 50 million parameters. You design the architecture, curate the data, and write the training loop. We want to see how much capability you can pull out of a deliberately small model. See the Rules page for exactly what the 50M cap includes and which benchmarks we score on.

Example directions: a domain-specific assistant for legal or cryptographic text; a model optimized for on-device inference; a small generative model trained on a narrow technical corpus without touching a commercial API.

Track 02: Applied (Medical Technology & Finance)

Build a data-driven system on real, empirical datasets that solves a problem in healthcare or finance. This means predictive models, diagnostic pipelines, or automated financial systems, backed by data engineering that turns raw input into something a person could actually act on.

Example directions: diagnostic software that processes device telemetry; analytics pipelines for pharmaceutical research; quantitative trading models driven by market indicators; automated risk scoring or fraud detection. All datasets must be public or de-identified, and prototypes must not be deployed on real patients or real money. See the Rules page.

Track 03: Open (General Technical Invention)

Build a functional prototype that shows technical novelty and cross-disciplinary engineering. Software, hardware, interactive media, generative systems, or any combination. No architectural limits and no resource constraints.

Example directions: IoT-connected physical infrastructure; a tool that changes how you work day to day; a generative AI product; a full-stack application aimed at an environmental or community problem.

Get started

1. Register on Devpost (July 11 to October 1)
Sign up here on Devpost. You must be registered on Devpost to submit, even if you signed up on our website first.

2. Build solo or form a team (1 to 6 members)
Work alone or assemble a team of up to 6. Once your team is set, pick the one track that fits what you want to build. Each team submits one project to one track.

3. Join the Discord (recommended)
Join the official GIBC V2 Discord to find teammates, ask the organizers questions, and get announcements: https://discord.gg/4y6RghaSt5

4. Submit before the deadline (October 1, 11:45pm GMT+8)
Submit all six required components on Devpost before the deadline. Late submissions are not reviewed.

Requirements

What to Build

Build a fully functional prototype, model, or technical system that directly aligns with one of our three challenge tracks: Track 01: TECH (LLM, 50M parameters), Track 02: Applied (Med/Finance), or Track 03: Open (General Invention).

Depending on your track, your project must meet these technical baselines:

Track 01: build and train a language model from scratch with no more than 50 million total trainable parameters, including embeddings and the output head. Full rules, reporting requirements, and the benchmark set are on the Rules page.
Track 02: build a system around empirical, real-world data science, predictive machine learning, or automated financial or medical pipelines. Datasets must be public or de-identified. See the Rules page.
Track 03: build a hardware schematic, advanced software stack, or an autonomous generative algorithm that breaks conventional structural boundaries.

The core requirement: strong submissions across all tracks show a working prototype, a complete end-to-end flow, handling of real-world complexity, and documentation clear enough that another developer could understand, run, and build on top of your work.

What to Submit

Your Devpost submission must include all six of the following to be eligible for judging:

01 | Project Description: a clear written explanation on your Devpost project page covering what you built, the problem it solves, and how it works under the hood.
02 | Source Code: a link to a public repository on GitHub, GitLab, Bitbucket, or Hugging Face, containing all project files. It must include a README with setup instructions, prerequisites, and usage guidelines so judges can run your solution.
03 | Demo Video: a 2 to 5 minute video on YouTube, Vimeo, or Youku, with the link in your submission. Unlisted is fine, private is not. The video must show your project in action and explain your methodology, in English audio or with English subtitles.
04 | Built With: a complete list of all technologies, frameworks, libraries, APIs, datasets, AI tools, and hardware used in your project.
05 | Team Information: all team members listed by real full name. Each teammate needs a Devpost account and must be added to the submission, or they will not appear and will not receive a certificate. This is required for verification and for issuing official GIBC V2 certificates.
06 | Screenshots: at least 3 high-quality images showing your interface, data pipelines, training loss curves, or technical setup in action.

Winners are announced at a virtual closing ceremony on October 8, 2026. Details on how to join will be posted on Devpost and in the Discord ahead of time. There is no additional presentation or preparation required after you submit. Judging is based entirely on your Devpost submission.

Hackathon Sponsors

Prizes

6 non-cash prizes

Gold (one per track)

3 winners

• Nord Security: 1-year subscriptions for NordVPN, NordPass and Incogni

• Saily: 1 GB of free data for Saily

• CodeCrafters: 2-year VIP membership.

• Art of Problem Solving: $25 coupon for AoPS courses.

• MeDo: 1-month MeDo membership, for projects built using MeDo.

• Starbit: October 17th, Jcard Verify × TABEI Cryptocurrency Flow Analysis Workshop - Basic Class

• Adaptation Labs: $1000 credits + 3 month plus plan subscription

• Official 1st Place certificate and global recognition.

Silver (one per track)

3 winners

• Nord Security: Nord1-year subscriptions for NordVPN, NordPass and Incogni

• Saily: 1 GB of free data for Saily

• Adaptation Labs: $1000 credits + 3 month plus plan subscription

• CodeCrafters: 1-month VIP membership.

• MeDo: 1-month MeDo membership, for projects built using MeDo.

• Official 2nd Place certificate and recognition.

Bronze (one per track)

3 winners

• CodeCrafters: 6-month VIP membership.

• Adaptation Labs: $1000 credits + 3 month plus plan subscription

• MeDo: 1-month MeDo membership, for projects built using MeDo.

• Official 3rd Place certificate and recognition.

Grand Champion

1 winner

• Featherless AI Grand Award: $300 in Featherless AI credits (awarded to the overall highest-scoring project among the three tracks),

• Official Grand Winner Certificate & Recognition.

All Participants

9,999 winners

• Official Certificate of Participation & Recognition,

• Featherless AI: 1 month of free LLM inference access during the hackathon,

• Devswarm: 1 month of DevSwarm Pro (Coupon code in Starter Kit)

• Interview Cake: Free coding/interview preparation course access,

• .xyz Registry: 1 free .xyz domain per team, free for the first year, 150 available on a first-come, first-served basis. Renewal after year one is paid at standard rates,

• MeDo Perk: 300 MeDo credits upon registering on MeDo during the event.

• Official GIBC partipcantation certificate

All Participants who Sign up For adaption Labs

9,999 winners

• You can Claim your $700 credits on Adaption Labs once you finish creating a account and Filling out on this form https://forms.gle/FRMCrg3akMVCoqfD9

⚠️You can only claim this prize before september 29

Judges

TBA
TBA

Mike
Starbit

Rohan Verma
CTO SurfSens

Eesha Tariq
Coding Instructor

Anshuman Khandelwal
LPL Financial

Deepika Sidana
American Express

Judging Criteria

LLM (Perplexity & Accuracy)
Language modeling quality on held-out validation sets. Lower perplexity and higher benchmark accuracy score higher.
LLM (Reasoning Performance)
Performance on logic, math, and commonsense reasoning tasks. Evaluated through standardized benchmarks.
LLM (Training Efficiency)
Compute efficiency, training speed, and memory footprint. Efficient architectures and training strategies rewarded.
LLM (Innovation)
Novel architecture decisions, creative data strategies, or unique training approaches.
LLM (Documentation & Demo)
Clarity of your README, reporting of parameter count, training setup, and benchmark results, and how well your demo video explains what you built.
MED / FINANCE (Innovation & Impact)
How novel is your invention? Does it address a genuine medical, scientific, or financial need with measurable impact?
MED / FINANCE (Technical Feasibility)
Does your prototype actually work? Quality of design, code, and engineering methodology.
MED / FINANCE (Rigor & Validation)
Quality of research, experimental design, data collection, and validation of your approach.
MED / FINANCE (Presentation)
Clarity of your demo, explanation of your invention, and ability to communicate its value to judges.
GENERAL INVENTION (Creativity)
Originality and boldness of the concept. How unique is the idea? Does it push creative boundaries?
GENERAL INVENTION (Execution)
Technical quality, polish, and usability. Does the prototype work well? Is the experience refined?
GENERAL INVENTION (Impact)
Potential to inspire, educate, or solve a real problem. Does the project leave an impression?
GENERAL INVENTION (Presentation)
Clarity and quality of the final demo. Can the team effectively communicate their vision?