Beevo

"From your mouth to the world!"

Comment

Inspiration

For millions of small business owners, branding is a nightmare. It's either expensive (hiring an agency for $15k+), slow (weeks of back-and-forth), or fragmented (juggling Canva, ChatGPT, and website builders). We realized that while AI tools exist for generating parts of a brand—like a logo or some copy—there was no "agent" that could act as a cohesive marketing department.

We drew heavy inspiration from Pomelli, Google Labs' AI marketing tool. Pomelli is fantastic for generating on-brand social assets and campaign ideas—it showed us what's possible when AI understands a brand's DNA. However, Pomelli is designed to assist a marketer. We wanted to replace the need for one entirely.

Beevo is the upgrade. It moves beyond content generation to autonomous execution. While Pomelli waits for your approval to post an Instagram story, Beevo is already building your landing page, researching your competitors, and refining your value proposition—all without you lifting a finger.

We wanted to build something that doesn't just "help" you work, but actually does the work for you. Our inspiration was to create a "Sovereign Virtual CMO"—an AI that can research your market, strategize your positioning, and execute a full landing page while you sleep using your brand's true identity. We wanted to move from "AI as a tool" to "AI as a workforce."

What it does

Beevo turns a 6-week branding process into 6 minutes. It is an autonomous marketing agent that:

  1. Talks to you via real-time voice (Gemini Live) to understand your vision.
  2. Researches competitors using live Google Search Grounding to find strategic gaps.
  3. Generates differentiated assets (logos, palettes, typography) that stand out from the competition.
  4. Builds & Optimizes a full landing page using "Watcher Agents" that visually analyze and iteratively improve the site.

How we built it

Beevo is built on a sophisticated multi-agent architecture powered by the Gemini 3 ecosystem.

The Brain: Gemini 3

We utilized specific Gemini models for their unique strengths:

  • Gemini 3 Pro for "Deep Think" strategic reasoning, ensuring every decision (like color psychology) is backed by logic.
  • Gemini Live API for the voice interface, enabling natural, interruptible conversations.
  • Gemini 3 Flash for high-speed generation of assets and copy.
  • Gemini Vision for the "eyes" of our Watcher Agents, allowing them to critique landing pages like a human designer.
  • Google Search Grounding to give the AI real-time awareness of the market.

The Body: Technical Stack

  • Frontend: React (Vite) + ReactFlow. We built an "Infinite Brand Canvas" to visualize the AI's thought process, making the invisible visible.
  • Backend: Node.js with a custom WebSocket architecture to manage the state between the user, the voice stream, and 6+ autonomous agents running in parallel.
  • The "Hands": We used Puppeteer heavily. One agent browses the web to "prune like an artist" (gathering logo inspiration), while others render the landing page code to simple screenshots for the Vision model to critique.

Challenges we ran into:

  1. Orchestrating Chaos: Managing state between a live voice conversation, real-time function calls, and 6 autonomous "Watcher" agents was difficult. We had to build a custom event-driven state manager to ensure the frontend didn't desync from the backend's "brain."
  2. Puppeteer in the Cloud: Getting a headless browser (Puppeteer) to run reliably in a serverless container (Google Cloud Run) was a significant DevOps hurdle, requiring careful Docker configuration and memory management.
  3. Voice Latency: creating a "magical" feeling voice interface meant shaving milliseconds off the response time. We had to optimize the WebSocket stream to handle audio chunks and function calls simultaneously without "stuttering."

Accomplishments that we're proud of

  • Thought Signatures: We're most proud of the "transparent reasoning" engine. Instead of just giving you a result, Beevo shows you why it made a decision (e.g., "I chose blue because your top 3 competitors use red").
  • The "Watcher" System: Seeing 6 distinct agents (Hero Watcher, Social Proof Watcher, etc.) collaboratively edit a landing page in real-time is mesmerizing. It feels like watching a team of ghosts working on your computer.
  • Live Market Intelligence: Beevo uses Search Grounding to pull real data, making its strategy genuinely useful.

What we learned

  • Transparency builds Trust: Users are much more likely to accept an AI's design choice if they can see the "Thought Signature" behind it.
  • Agents need Guardrails: Autonomous agents are powerful but can go off-rails. We learned the importance of a "Human-in-the-Loop" Command Center to guide the AI without stopping it.
  • Multimodal is the Future: Combining Voice (Input), Vision (Feedback), and Code Generation (Output) creates an experience that feels fundamentally different from text-based chat.

What's next for Beevo

We're just scratching the surface of autonomous marketing. Here's our roadmap:

  1. Full Funnel Automation: Beyond just landing pages, we're building "Ad Watchers" that generate and optimize Facebook/Instagram ad creatives that drive traffic to the landing page.
  2. Multilingual Expansion: Using Gemini's advanced translation capabilities to automatically localize brands for global markets with a single click.
  3. eCommerce Integration: Moving beyond lead-gen pages to full Shopify store generation, with agents that write product descriptions and optimize pricing.
  4. Brand Bible Generation: Automatically compiling all the generated assets into a professional PDF Brand Guideline document that businesses can hand to human designers.

Built With

Share this project:

Updates