Membrane

Inspiration

Our mother frequently phone calls my grandfather, who is a very spiritual person. He frequently tells our mom that he's praying for her or having rituals done on her behalf. Our grandfather is hard of hearing and nowadays easily trusting, and we realized that if his data was exploited by someone pretending to be my mom, he could easily be taken advantage of. At the same time, he's not really the type to care about cybersecurity, as that's what our uncle normally does and my grandfather trusts him on that. This made us realize that the solution needs to be one that doesn't complicate the experience for the elderly or give up their privacy while still maintaining the strength of proper cybersecurity methods.

What it does

Membrane is an orchestration and guardrail layer that sits between senders and receivers, acting like a cell membrane. A suite of agents divides the work among themselves: recovery, sender authentication, and receiver authentication, and adaptability. Each of them relies on separate, unique, interchangeable signals that don't leak identity or biometric data. The agents learn to verify those signals, manage the handoffs between one another, and adapt to the receiver's particular needs. Where a vulnerability exists, they harden protection around it rather than just flagging it, and when they block a malicious actor they explain why. If the agents can't resolve a case on their own, a trusted individual is brought in as a final layer of verification. The tool is also meant to use its adaptive learning to provide recovery and accessible authentication for its users as well through local verification methods that are less likely to leak important biometric and personal data.

How we built it

We created this through a mixture of vibecoding with AI tools like Claude and ChatGPT, as well as with our own coding knowledge by hand. We did research on the different packages and software needed to form our tech stack (mainly using Claude) and were able to integrate them to achieve our goal.

Challenges we ran into

The main challenges we ran into were in the ideation process and somewhere down the build process where it came to building our MVP and demo. Ideation took the longest time, as we wanted to make sure that we really found a solution and a problem to solve that were real. We ran into technical bugs later on and took some time to fix them, which was stressful.

Accomplishments that we're proud of

We think that our idea is an interesting one, and we're proud that we were able to translate that into reality- at least somewhat. We also realized how well we work together as a team under pressure and how naturally we divide up responsibilities.

What we learned

We learned the importance of time management due to the short timeframe we had available for building Membrane. We also learned several ideation methods and angles that we could think from that we didn't have before, which we think will help us in the future. We also learned more things about the tools we used and their limitations as well as their strengths.

What's next for Membrane

The next steps for Membrane are to flesh out its agentic toolkit and strengthen the actual privacy and security features. We also want to further the MVP to actually serve its various use cases and sit in a platform that is accessible on all sorts of ecosystems.

Built With

• claude
• react
• simplewebauthn
• supabase
• twiliosms
• typescript
• vite