DeepGuard

Inspiration

DeepGuard has been on my mind since 2022 when my former boss and I created a DeepFake for security testing. In early 2024, I listened to the CEO of a Fortune 500 speak about how senior leaders at the company all fell victim to a test DeepFake scam in which the testers took a 10 second clip of the CEO speaking and then proceeded to call those senior leaders with an AI generated audio message asking them for sensitive files, wire money, or do something that went against security protocols. They did this to ~20 people and every single one of them complied with the DeepFake's request(s).

I personally feel like we need this technology/solution/tool to combat these new types of fraud since something like 98% of humans cannot detect a DeepFake now that they have become so sophisticated. These attacks are especially insidious when done to vulnerable segments of the population, handicapping many people who are not tech savvy or may not even know what a DeepFake is. Beyond the societal implications, I also firmly believe that this is the type of technology/solution/tool that will be very much needed for the security of government(s) and citizens.

What it does

DeepGuard is an enterprise-grade B2B platform that provides proactive protection against synthetic media and AI-generated content across multiple channels. This solution empowers businesses to detect, verify, and respond to DeepFakes, manipulated images, synthetic audio, and AI-generated text before they can cause reputational damage, security breaches, or operational disruption. The platform will offer real-time monitoring of incoming media across email, video conferencing, customer uploads, and internal communications.

How we built it

We used Claude to generate detailed prompts to put into Lovable, used Lovable to generate the code to build the site, and connected Supabase to use for the backend.

Challenges we ran into

Lovable is a horrible, awful, terrible, blood-sucking platform to use. We used Claude and provided it with supporting documents (PRD, Style Guide, etc.), and still burned through ~130 credits because Lovable kept breaking things or simply didn't perform the tasks it was asked to do. Even with the Buildathon instructions/guide provided, Lovable did some very peculiar things like completely ignoring certain parts of clearly defined prompts, stopping in the middle of a build to ask if I wanted to continue, and charging credits whilst in "chat" mode (which it isn't supposed to do). The AI would claim to have found "the exact" cause of a problem, but never seemed to be able to solve it on the first try despite confidently stating that it was going to do so. I'm not sure if this is due to bad spec-driven prompting, but it sure felt like Lovable was behaving the way it did to get me to use more credits (spend more money).

Accomplishments that we're proud of

Neither of us had used Lovable, Supabase, or Resend before, and considering how user-unfriendly Lovable is, I think it's pretty impressive that we managed to get a (mostly) working app, even if it's a bit rudimentary.

Getting the damn thing done! I've certainly worked against fast-approaching deadlines before, but it was still a lot to do (listen, learn, network, and build) in the given timeframe even with the deadline extension!

What we learned

• The power of spec-driven development over prompt engineering.
  
• Vibe coding isn't a magical solution that allows anyone to create an app in seconds like the marketing claims. You still need to understand principles of software engineering (and you really need to at least have basic coding skills) to use a vibe coding platform in order to produce a clean, functional app.
• Several great tips on product refinement, business strategy, and pitch strategy.

What's next for DeepGuard

Research and iteration are the most immediate needs for DeepGuard, followed closely by product refinement, technology selection (for scaling), market research, and business strategy refinement. And somewhere during all of that, a team of passionate builders/developers/professionals needs to be formed.

Built With

• claude
• github
• lovable
• resend
• supabase