Flair

Architecture: Three ADK agents. Four Gemini capabilities. One pipeline — from brand discovery to campaign production, on Google Cloud Run.
Full campaign packag- storyboard, taglines, platform strategy. Ready to post.
Flair Campaign Studio - four modes, one pipeline. Start with Discovery to build your Brand Brain.
Discovery Agent in action - Gemini Live API, real-time, barge-in supported
Three campaign directions & hero images - one request, one interleaved stream.

The problem nobody talks about

Every AI content tool on the market does the same thing: you type a prompt, it generates something that looks fine and says nothing. The output is polished, generic, and forgettable — because the AI knows nothing about your brand, your audience, or why any of this matters.

We asked ourselves: if you hired a great marketing consultant, would they start by making you an Instagram post? No. They'd spend the first week just listening. Understanding your product, studying your competitors, learning how your customers actually talk and think. The creative work comes last, not first — and that's why it's good.

What Flair does differently

Flair is an AI marketing strategist that does the homework before it does the work.

It has three parts:

Discovery — A live voice conversation (powered by Gemini's Live API) where Flair acts as your marketing consultant. It reviews your materials, researches your market, and asks the questions a real strategist would ask. In about two hours, it builds a deep understanding of your brand, audience, and competitive landscape — what we call the Brand Brain.
Studio — Where assets get produced. But unlike every other tool, Studio doesn't guess. It draws on everything Discovery learned to produce creative that's strategically targeted, emotionally precise, and built to perform — not just look pretty.
Refine — The part that makes everything smarter over time. It ingests real performance data, analyzes what's working, spots trends, and feeds those insights back into Studio. The more you use Flair, the better it gets.

Why Gemini

This project exists because of Gemini's Live API. The Discovery conversation — where a marketing strategist talks with you in real time, reacts to what you say, presents research, and adapts — that's not possible with a text prompt or a chatbot. It requires bidirectional voice, multimodal understanding (we analyze images, documents, websites), and the ability to hold a long, nuanced conversation with memory. Gemini's Live API with session resumption and context compression makes that real.

We also use Gemini for image generation (campaign visuals, storyboard frames), structured content generation (strategy briefs, audience portraits), and competitive research synthesis.

What we learned

The hardest part wasn't the technology. It was designing a conversation that feels like working with a real person — one who comes prepared, respects your time, and occasionally tells you something about your own business you hadn't considered. Getting the pacing, the tone, and the interaction patterns right matters more than any API call.

Challenges

Keeping a coherent, memory-rich conversation alive across Gemini Live API's session boundaries (10-minute WebSocket limits, 15-minute context windows) required a layered resilience strategy: context window compression for long sessions, session resumption tokens for seamless reconnection, and structured function calls that persist key findings so nothing important is ever lost — even if the connection drops.