Irabu

Inspiration

Back in Africa, I went to public school, which meant no access to the resources students at private schools took for granted: no advanced books, no dedicated programs, no real learning tools. I remember my school receiving a donation of iPads from UKAID, meant to give us access to modern learning technology. They were never used. Not once. One day I asked my teacher why. He told me the truth: the technology was too heavy for most of the kids there, and for the teachers too. It wasn't an excuse, it was the reality. Many students came from low-income families, and in a place like that, poverty means no prior exposure to a computer, a smartphone, or any technology at all. Adapting to something like an iPad wasn't a small ask for them, it was a leap nobody had prepared them to make. And most of our teachers didn't have the technology literacy to bridge that gap either. So the devices sat unused, a well-intentioned donation that solved nothing, because the problem was never access to hardware, it was access to something people could actually use. After I graduated high school, I volunteered at a local NGO, and I watched the same failure play out in a different form. The organization's mission was clear, its people worked hard, and the problems it was solving in the community were real and urgent. But the tools weren't there. We wrote weekly plans on large sheets of paper and stuck them to the wall. We kept records on more paper, until there was so much of it that retaining information became nearly impossible. Nothing was organized, not because anyone was careless, but because the system itself had no way to hold what the organization was trying to do. Limited funding meant limited tools, and limited tools meant an organization full of good intentions running on paper and memory. At the time, I couldn't fix either of those problems. Now I can.

What it does

I decided to build iRABU.works with over 30+ agentic tools and workflows to solve these two problems, because at moment i wasn't able to do so, but right now, I have the capability to do so. iRABU helps underesources groups or communities to have access to the productivity tool where the organization can no longer write records on papers, weekly plannings on big sheets and dispose at the end of the week, but will have a specific program designated for team collaboration, creating and tracking tasks and goals, managing finances, create workflows, manage meetings, advanced data analytics and above all, all these capabilities can be executed in a chat interface with agentic operations so that teams can focus on being productive. And funding won't limit organizations from using our platforms, we are using Solana technology, so that donors can donate specifically for a designated organization to have access to the platform. Everything happens in blockchain so that donor's funds can go straight to platform costs and not anything else. With iRABU even teachers can uplod topic notes, and through AI chat, students can learn and access topic information grounded to what knowledge resources the teacher has uploaded, not a general generative AI output that can sometimes give generalized knowledge, which ends up confusing students just like in previous iPad story.

How we built it

iRABU.works is an AI-powered productivity platform, over 30 agentic tools and workflows, all reachable from a single chat interface, built specifically for under-resourced organizations: public schools, NGOs, nonprofits, farmer co-ops, small businesses, and community groups running on too little funding and too few staff to run a normal enterprise software stack. Instead of paper weekly plans stuck to a wall, iRABU gives a team one place to create and track tasks and goals, manage meetings, run financial workflows, analyze data, and build automations, all through natural conversation with an assistant that can actually act, not just answer questions. Ask it something about the organization's own documents and it answers with citations back to the source. Ask it to chart a set of numbers, draft an email, schedule a meeting, or log a task, and it does it in the same conversation. The knowledge that used to live on paper now lives in one searchable, structured place a whole team can share. And because funding is the other half of the problem, iRABU doesn't assume an organization can just pay for software. We built a sponsorship model on Solana: a donor can fund a specific organization's platform usage directly, and every action the assistant takes, every answer, every task, every drafted email is metered and settled on-chain as a public, verifiable ledger. The donor sees exactly what their funding paid for, without ever seeing what anyone actually asked. The money can only ever cover platform costs; it can't be withdrawn or redirected anywhere else. It's the same transparency an NGO owes its own donors, applied to how the tool itself gets paid for.

Challenges we ran into

The hardest bugs weren't in any single query, they were in the interaction between systems, and we only found them by testing against a real, live database instead of trusting the SQL on paper. One RLS policy checked a sensitive-source flag through a subquery into another table, but that table had its own RLS policy, which silently hid the very row the check needed to see, flipping "denied" into "allowed" for exactly the people it was supposed to block. A second bug let a user approve themselves for a completely different organization's sensitive data, because an authorization flag lived globally on a user profile instead of scoped per organization. A third meant our audit log would have silently never recorded a denied access attempt, because the check relied on a session identity that doesn't exist for backend service calls. We found, understood, and fixed all three before calling the feature done. Third-party APIs had gaps their docs didn't mention. A memory-storage API silently skipped saving anything unless it was also generating a paid reply. A sandbox payments API's balance field turned out to be effectively read-only, so we had to build our own ledger on top of it. A devnet faucet rate-limited real traffic mid-build. Each one took direct, empirical testing, not just reading documentation to actually understand. And partway through, we ran out of real Anthropic API credits, which is its own kind of humbling bug report to get mid-hackathon.

Accomplishments that we're proud of

Everything here is live-tested against real APIs, real Solana devnet transactions you can verify on-chain, real Persona sandbox inquiries, real embeddings, real streamed chat, nothing mocked or faked for a demo. We're proud that the sponsorship model isn't a billing gimmick; it's built to directly answer the funding fragility from the Inspiration story, transparently and without giving anyone a way to divert the money. And we're proud that when we found real security bugs in the access-control system, we didn't ship around them, we traced each one to its root cause, fixed it, and verified the fix live before moving on.

What we learned

That powerful technology nobody can actually use isn't a solution, it's the iPad donation all over again, just in software instead of hardware. That security for an AI agent has to live in the infrastructure, the database, the RLS policy and not in a prompt instruction, because a prompt is a conversation the user is already inside of. And that identity and authorization are two different questions: proving someone is a real human is not the same as proving they're allowed to see an organization's most sensitive data, and conflating the two is exactly where access-control bugs hide.

What's next for Irabu

Moving Persona verification from a client-confirmed check to full webhook-based confirmation, so verification status doesn't depend on a browser tab staying open. Extending credential verification beyond identity, proving someone is a licensed clinician or an authorized staff member, not just a real person. A real pilot with an actual under-resourced school or NGO, pressure-testing the platform against their real workflows instead of a synthetic demo. And a path to Solana mainnet once real donor funds are ready to move, carrying forward the same public, spend-restricted ledger we've already proven on devnet.

Built With

• anthropic-claude
• backboard
• capital-one-nessie-api
• elevenlabs
• next.js
• persona
• pgvector
• postgresql
• react
• resend
• row-level-security
• solana
• supabase
• tailwind-css
• typescript
• vercel
• vercel-ai-sdk
• voltr
• voyage-ai
• web3.js