Your guide to the 48 hours
At a glance
The challenge & rules
Build whatever fits the problem: an app, an assistant, an agent, a tool, a pipeline. It doesn't have to be an agent. The problem is yours to choose across everyday life, government services, business, education, transport, culture or industry.
The one requirement: build with Hugging Face
Your project has to be built with Hugging Face. That can mean our open-source libraries, our platform, or both. For example:
- Use the open-source stack — Transformers, Diffusers, TRL and friends
- Fine-tune a model on Jobs
- Call models through Inference Providers
- Stand up an Inference Endpoint (vLLM, SGLang or custom)
- Use Sandboxes, or plug Agent Skills into your build
See the resources section for links to all of these.
Fine-tuning is optional
Fine-tune only if your project needs it. Plenty of strong projects will just use inference. If you do fine-tune, smaller models are cheaper and faster to work with, but Jobs gives you access to powerful GPUs if you want to go bigger.
What to build
- A working demo as a Hugging Face Space people can try
- A 2 to 3 minute demo video, your backup if a demo breaks on the day
- If you fine-tuned a model, push it to the Hub with a model card
Schedule
48 hours, from Thursday 5:00 PM (Doha time). The moments that matter for builders:
| When | What |
|---|---|
| Thu, 5:00 PM | Kickoff and challenge reveal. Building starts. |
| Fri | Full build day, workshops and a fireside chat. Note: venue pauses 11–1 for Friday prayer. |
| Sat, 2:00 PM | Submission deadline, hard |
| Sat, afternoon | Pitches, judging and awards. |
Account & setup
Before the event (everyone)
username/my-model). Confirm your email, because without it you can't use credits or claim the coupon.pip install -U huggingface_hubDay 1: team credits
$100 per team, one shared pool. Nominate one team captain who holds it.
hf auth login --token $TOKENCredits & compute
- Billing is per-minute, only while things run. Covers Jobs, Inference Endpoints and Providers in one pool.
- Check prices before you launch:
hf jobs hardware· Jobs guide - Your $100 covers fine-tuning and your demo, with no top-ups, so budget across the weekend.
Jobs survival tips
hf jobs uv run --flavor a10g-large --timeout 3h train.py- Save to the Hub from inside the job (
push_to_hub=Trueor checkpoints), because the job's disk is wiped when it ends. - Watch and stop idle jobs:
hf jobs psandhf jobs cancel <id>. Check before you leave at night.
Ways to use Hugging Face
Pick whatever fits your project. A few starting points:
- Fine-tune a model on Jobs: Jobs guide · beginner fine-tuning with Unsloth
- Call hosted models via Inference Providers
- Stand up an endpoint (vLLM, SGLang or custom) with Inference Endpoints
- Run code safely with Sandboxes
- Plug skills into your agent with Agent Skills (and HF Skills)
- Go local (nice to have, not required) with llama.cpp
How to submit
Where your work lives
There's no team org. Your Space (and your model, if you fine-tuned one) sit under your team captain's account, the one holding the credits that your tokens are tied to. Name them with your Team ID so they're easy to spot:
- Space:
captain-username/team-NN-<short-name>-app - Model (if you fine-tuned one):
captain-username/team-NN-<short-name>
Use lowercase and hyphens, add the tag qai-hf-hackathon-2026, and if you have a model, link it from the Space. Then paste the links in the sheet.
Your submission is complete when
- Your Space is live under the captain's account, tagged, and running
- If you fine-tuned a model, it's on the Hub with a card and the tag
- Demo video (2 to 3 min), link viewable by anyone
- Your row in the submission sheet filled and marked Submitted
- Everything built during the event
Optional: write a short blog post about what you built and drop the link in the "Blog URL" column of the sheet. A nice way to share your work, and we love to see it.
Open the submission sheet →How you're judged
| Criterion | What we look for |
|---|---|
| Impact & relevance | A real, worthwhile Qatar or Middle East problem, solved well |
| Use of Hugging Face | Your project is genuinely built with Hugging Face |
| Technical execution | It actually works end to end |
| Efficiency | Smart, resourceful use of your compute budget |
| Demo & pitch | How clearly and compellingly you show it off |
FAQ
Does it have to be an agent?
No. Build whatever fits the problem: an app, an assistant, an agent, a tool or a pipeline. The one requirement is that it's built with Hugging Face.
Do we have to fine-tune a model?
No, fine-tuning is optional. Many strong projects will just use inference. Fine-tune only if your project needs it.
Does it have to be about Qatar?
It must be grounded in Qatar or the wider Middle East. Projects that aren't clearly contextualized to the region will be disqualified.
Can we use a frontier model (GPT, Claude, etc.)?
You can use frontier models around the edges, but your project needs to be genuinely built with Hugging Face, and that's what we score.
Is $100 really enough?
Yes, if you're deliberate. Check pricing before big runs, and lean on inference or efficient fine-tuning. No top-ups, so plan it.
Can we get more credits mid-event?
No. Working creatively within the budget is part of the challenge.
Help & contacts
Hugging Face engineers are on-site as mentors across all 48 hours, so grab them. The event channel (shared with your acceptance email and at kickoff) is your home base for setup, credits, Jobs and submission questions, with office hours each day.
Venue: Nestwork & Scale 7, Abdulla Bin Thani St, Doha. Questions: info@trustqai.com
Scan the QR, or tap the button on your phone, to join the Hackathon01 group — announcements, Q&A and updates during the event.
Join on WhatsApp →Quick links: full agenda · submission sheet