Inspiration
Real estate teams sell possibilities before buyers can see them. Empty rooms, dated kitchens, rough listings, and unfinished spaces often need imagination, but imagination is slow to package. Seen was inspired by that gap: helping agents, brokers, builders, and property brands turn a raw property photo into a clear visual story.
What it does
Seen is a creative real estate AI agent that helps transform property photos into finished listing media. A user uploads or pastes a property image, describes the buyer, mood, room, and goal, then the agent asks clarifying questions before generating staged images or short video concepts. It also tracks credits, project history, approvals, and generated outputs inside a workspace.
How we built it
We built Seen with Next.js, Google ADK for the agent workflow, Gemini for reasoning, MongoDB for users, briefs, credits, history, and generated media records, Better Auth with Google OAuth for sign-in, Polar for credit-based billing, and Phoenix Cloud for observability. The media generation flow is wired through agent tools so image and video work can be traced, stored, and shown back in the dashboard.
Challenges we ran into
The hardest part was making the agent feel real instead of like a static form. It needed to ask useful questions, wait for confirmation, use the uploaded property image, call the right media tool, and return honest output states. We also had to wire auth, credits, MongoDB persistence, checkout, and observability without letting the product become noisy or technical for the user.
Accomplishments that we're proud of
We built an end-to-end workspace where a real estate user can sign in, upload or paste a property image, chat with an agent, confirm a brief, spend credits, and receive generated media tied to their project history. We also added Phoenix tracing around the agent and tool calls, so the system can be debugged and improved instead of operating like a black box.
What we learned
We learned that agent UX depends as much on restraint as capability. The agent should not rush into generation; it should gather the right context, explain the direction, and wait for confirmation. We also learned that observability matters early, because media generation involves multiple moving pieces: user intent, tool calls, provider latency, credits, and stored outputs.
What's next for Seen
Next for Seen is stronger property-specific editing, richer video generation, team workspaces, approval flows for broker teams, saved brand styles, listing copy generation, and production-ready credit plans. We also want the agent to learn from each brand’s past projects so future property concepts feel more consistent, faster, and more useful.
Log in or sign up for Devpost to join the conversation.