Builder guide · Read before you code
Talk to the problem before you build the solution.
Great hackathon projects fix one narrow, real bottleneck. This guide helps you research a real civic problem, shrink it to something you can finish, and ship a working tool by Sunday.
Step 1
Pick a problem small enough to finish
- Question 1
Who is it for?
Name the specific person (an intake coordinator, a pantry volunteer, an after-school parent), not just "nonprofits."
- Question 2
What do they do today?
Map the manual, messy steps they take right now: spreadsheets, paper forms, back-and-forth phone calls.
- Question 3
What breaks?
Find where the delay, confusion, or lost data happens.
- Question 4
What would "better by Sunday" look like?
Define the smallest working slice that saves someone 30 minutes or clears a backlog.
Too big vs. just right
- Too big
An end-to-end AI platform to solve housing insecurity across NYC.
Too broad, impossible to test, no specific user.
Just rightAn intake assistant that reads tenant eviction notices and flags court deadlines in plain language.
One input, one output, immediately useful to a legal aid screener.
- Too big
A multi-agent system that runs a nonprofit's whole volunteer schedule and donor database.
Brittle, needs complex permissions, hard to demo safely in four minutes.
Just rightAn SMS auto-responder that answers the top 5 volunteer questions in English and Spanish.
Clear scope, checkable answers, easy to demo live.
- Too big
A self-navigating robot that fully maintains a community garden.
Hardware and weather take weeks to calibrate.
Just rightA tabletop robot arm with computer vision that sorts dry seed pods from viable ones.
One physical sorting task with a clear pass or fail.
Step 2
Research the problem with public sources
You don't have to guess what NYC community groups need. Start here.
- The nonprofit’s websiteActive programs and the communities they prioritize.
- Annual reportsGoals, milestones, and funding limits leadership actually tracks.
- IRS Form 990 on ProPublica (opens in a new tab)Staff size and budget, so you know what tools they can realistically maintain.
- Their Instagram and LinkedInEvent posts and volunteer call-outs show real headaches happening right now.
- NYC Open Data (opens in a new tab)Free datasets on city facilities, permits, social services, and food retail.
- NYC 311 data (opens in a new tab)Service requests by neighborhood and zip code show recurring friction.
- US Census (ACS) (opens in a new tab)Languages spoken at home and household income, so your interface fits the people using it.
Step 3
Talk to real people (10 minutes is enough)
Ask a mentor or nonprofit partner these five questions.
- 1
What repetitive, manual task eats up the most hours of your staff’s week?
- 2
When that workflow backs up or fails, what happens right away?
- 3
What software or workaround do you use now, and where does it fall short?
- 4
If an assistant took over just the first step of this task, what would that look like?
- 5
If we hand you a working prototype Sunday afternoon, what one thing must it do for you to say “this helps”?
Client data: never collect, record, or put confidential client details, medical notes, case numbers, or real identities into an AI model.
Track
Physical AI for Good, with Viam
This track pairs software with physical motion using Viam robot arm kits. Physical AI teams get the hardware starting Friday at 11am. Here's what an arm can realistically do in two days:
Pick and place
Move specific items from a drop point into a collection bin.
Color or shape sorting
Use a fixed camera to identify an object and route it to the right tray.
Physical status checks
Check whether a container is empty, upright, or misaligned.
Hardware safety
- Keep fingers, hair, badges, and loose clothing clear of the arm’s joints and motors.
- Secure the arm’s base to a flat, level surface before powering it up.
- Keep your workspace clear of loose cords, food, and open drinks.
- Never leave a powered, moving arm unattended.
Setup guide:Coming soon
Rules
Data and privacy
Public or synthetic data only
Train, test, and demo only on public datasets or mock data you make yourself.
Zero personal information
Strip real names, phone numbers, addresses, and case files before any text goes into a prompt or API call.
Mask user inputs
If your app takes submissions, anonymize them on the device before sending them to a server.
Protect your keys
Never commit API keys, webhooks, or credentials to a public repo. Use environment variables and .gitignore.
Partner tools
Viam
Credit steps comingPhysical AI robotics platform and the robot arm kits.
Vonage
Credit steps comingCommunication APIs for voice, SMS, and messaging.
Submit
What a strong submission looks like
Real problem
It tackles a validated civic or nonprofit bottleneck, not an imaginary one.
It actually works
It runs and demos real functionality, not a clickable mockup.
Safety and privacy
It guards against bad outputs, protects private data, and uses AI responsibly.
Clarity
The team can explain what it does, who it serves, and how it works in plain English.
Every team submits a repo link, a demo video, and this short writeup. See the submit section for the deadline.
### 1. Problem What specific bottleneck does this solve, and what is broken about how it's done today? ### 2. Who it helps Who is the exact user, nonprofit, or community member using this tool? ### 3. What we built What does the tool do, how does it work, and what powers it? ### 4. What's next What would it take to pilot this with a partner organization next month?
Stuck?
Get help
Discord
Ask technical questions, find teammates, and get scoping feedback in the event channels.
Mentors on site
Technical and nonprofit mentors walk the floor to help you get unstuck.
Anyone with a volunteer name tag
Power strips, directions, hardware help, or schedule questions: ask a volunteer.
Schedule and venues
Times and addresses for all three days live on the main hackathon page.