Loading organizations…
Loading organizations…
Mangrove helps people go from caring about AI safety to producing real AI safety work. Finishing a project is the fastest way to build skills and context when fellowship spots are scarce. It gives you something concrete to show, and it puts more hands on problems the field would otherwise drop.
Most people who care never get that far. Getting started takes more energy than listening to one more podcast. The rest of life keeps winning the fight for your time. And when the people around you find worry about AI risk strange, it is hard to stay on the path.
Mangrove takes that friction away. Fill in one common application and say when you are free. An AI model then suggests projects and events you might like. Applying to one takes a sentence or two, because the common app already holds the rest. You only get matched into teams whose free hours overlap, so the scheduling that kills most group projects is gone. Project leaders pick their team from the applicants and launch, and projects are open all year, not just in a fellowship season. If you would rather work on your own, join an accountability sprint: pledge to weekly co-working sessions and demo your output at the end. Every team gets a private workspace set up automatically, with a shared drive, a chat channel, group scheduling, a task tracker, and optional meeting notes that update it. A track record certifies reliable attendance and flags no-shows, and peer reviews spotlight quality work.
When a project is done, you have output people can inspect, not just a line on a resume. Your teammates review your work, and you can add the artifacts, the reviews, and your profile to AI safety hiring boards. That gives employers and grantmakers a real basis for jobs and funding. Promising projects can keep going: teams can take on more members, and grantmakers can fund the next stage.
Mangrove also runs hackathons: short, focused sprints where teams build inside a fixed window and review each other against criteria published before anyone starts. Sign up alone and we match you with a team before kickoff.
To get involved, create an account at try.mangrove.one, fill in the common app, and join a project, a sprint, or the next hackathon. Organizations can list themselves in the directory and host events.
Mangrove's provides a scalable solution to AI safety upskilling, which is currently bottlenecked by fellowship capacity. Through weekly hackathons and monthly project tournaments, Mangrove creates a dopamine-rich environment for people to upskill, showcase their competencies, and get hired by AI safety employers. People start their own AI safety projects during hackathons, receive peer feedback each week as they compete for monthly awards, and in the process, develop a portfolio that employers can browse. For employers and grantmakers, people's inspectable output on Mangrove provides signal on who is promising. For strategists, Mangrove's hackathon software makes it easy to kickstart work on neglected issues. And for individuals, Mangrove makes it easier to for people to find their place within AI safety.
Mangrove: Matchmaking for AI Safety
Closed Aug 25, 2026
Software to match people with shared AIS interests into teams that complete projects, increasing the odds that people beyond scarce talent pipelines upskill, develop credibility, and turn their ideas into output.
$17.5K minimum · $33.5K ideal
We’re running a pilot of 56 people live at [www.mangrove.one](http://www.mangrove.one) and are discovering that providing peer accountability via online matchmaking is effective at motivating people to produce AI safety output rather than just reading about it. In the next month, we would like to 1. Scale these findings to onboard a beta cohort of 500 people. 2. Host hackathons on Mangrove to stress-test Mangrove’s infrastructure, so that Apart Research can feel confident about using Mangrove to facilitate its hackathon matchmaking. 3. Build out a track-record system that compiles project outputs, attendance records, and peer reviews. This would both improve the quality of Mangrove's matchmaking for repeat participants and create a datasource that employers and grantmakers can use to decide who to extend opportunities to. Our team has been self-funded since April and currently includes - Zach Hsu (Stanford, BCG) - Jayani Srinivasan (UC Berkeley, Apart Research hackathons, Blue Dot) - Josh Peng (AI Safety Camp, Blue Dot, 2x Mangrove participant)
Joining a peer-group to upskill into AI safety increases the odds that newcomers contribute to the field. However it rarely happens. Most people interested in AI safety default to exploring alone, or worse, returning to life as usual. Finding peers with shared interests, availability, and motivation to collaborate for a sustained period of time is difficult. By removing this friction · via software that matches people into volunteer teams · people gain motivation to upskill outside of structured fellowships. This enables the talent pipeline to expand beyond the capacity of AI safety's MATSs, Generators, or Horizons. That's what Mangrove does. Members share their interests and upload their weekly availability, then post project ideas or apply to existing ones. When enough people commit to an idea and their schedules overlap, Mangrove forms the team and provisions everything it needs to start: Slack, Drive, and suggested meeting times. Since people collaborate on one shared deliverable, skipping a meeting means letting real people down. This keeps people committed even when life events vie for people’s time…
Minimum ($17,500): - Hackathon prize pools (1.75k x2): $3,500 - Project compute fund: $3,000 - Development and infrastructure: $9,500 - Beta cohort outreach: $500 - Legal: $1,000 Ideal ($35,000), everything above plus: - Third hackathon prize pool: $1,750 - Project compute fund, expanded: $3,000 - Community manager stipend: $3,000 - Track-record system, deeper: completion diagnostic (never-started vs stalled-after-kickoff), full track-record rollup: $4,250 - Growth, expanded: $2,000 - Infrastructure headroom: $1,000
Fundraising data from grantmaking.ai, updated .
View on grantmaking.ai