Inspiration
Most influencer tools focus on posting more content, not better decisions. We wanted to build something that could survive in the long term. The façade was designed around the idea that a digital brand should understand its audience, its resources, and the moment it operates in. By combining fast reasoning with high-quality media generation, Façade behaves less like a bot and more like a manager. It is not just a content creator; it is a self-managing digital brand.
What it does
Façade is an autonomous influencer platform built around decision-making first and content second. Users define a persona with a backstory and goals, and Façade manages the rest.
Strategic Trend Decisions with Gemini 3 Flash
Façade continuously evaluates emerging trends using Gemini 3 Flash. Each trend is scored for its likelihood to perform well, allowing Façade to decide whether a trend is worth acting on at all.
High Quality Video Creation with Veo 3.1
Only trends with strong potential trigger Veo 3.1 Fast, which Façade uses to generate cinematic, scene-based videos. This ensures that high-effort video creation is reserved for moments that truly matter.
Persona Evolution Powered by Gemini
Gemini 3 Flash monitors audience feedback over time. When engagement drops or sentiment turns negative, Façade uses Gemini to analyze why the shift occurred and adjusts its interests and tone to regain engagement.
Real Time Generation
Rather than relying on fixed schedules, Façade reacts to real-world events as they happen, using Gemini 3 Flash for fast reasoning and Veo 3.1 for rapid, high-fidelity visual output.
Long Term Memory with Context Caching
Façade uses Google Cloud Context Caching to store its full persona history and goals. This allows Gemini 3 Flash to reason over thousands of background-context tokens at a fraction of the usual cost.
How we built it
Frontend
Built with Next.js, React, and Tailwind CSS, Façade includes a Mission Control Dashboard that shows the current reasoning state, trend confidence, and audience mood in real time.
Backend
A FastAPI server runs a custom event loop that orchestrates perception, reasoning, and action.
Gemini and Veo Integration
- Gemini 3 Flash is used for trend evaluation, sentiment analysis, persona decisions, caption writing, and output verification.
- Veo 3.1 Fast is used for high-quality video generation when Gemini determines the opportunity is worth the cost.
Database
SQLite stores persona memory, sentiment history, and post metadata.
Challenges we ran into
Balancing speed, quality, and cost was our biggest challenge. We initially generated videos too aggressively, which led us to rely more heavily on Gemini 3 Flash to filter trends before invoking Veo 3.1. We also encountered multimodal verification issues due to SDK version mismatches, which required careful alignment across the Gemini and Veo pipelines. Finally, we learned that deeper personas were required to fully take advantage of context caching, forcing us to expand our persona backstories for better long-term consistency.
Accomplishments that we're proud of
We successfully built a system that controls Gemini 3 Flash when Veo 3.1 is used, rather than blindly generating media. Watching Façade decide not to post, or to pivot its personality based on audience feedback, demonstrated real strategic behavior. We are also proud of achieving nearly a 90 percent reduction in ongoing context costs through effective use of Gemini context caching.
What we learned
We learned that Gemini is most effective when used as a decision-maker, not just a generator. The real innovation was letting Gemini decide whether something should be created before asking Veo to create it. We also gained hands-on experience designing fast, reliable multimodal pipelines by working with Gemini and Veo.
What's next for Façade
Next, Façade will integrate directly with the Instagram Graph API for live comment analysis and automated posting powered by Gemini 3 Flash. We also plan to enable rapid response video creation, allowing Gemini to detect breaking events and trigger Veo 3.1 to produce fully rendered video content in under 60 seconds.
Log in or sign up for Devpost to join the conversation.