-
-
Landing Page
-
Domino's product placement
-
Purchase option for on-screen products
-
Side-by-side of two user's different viewing experience
-
System architecture
Inspiration
Streaming has become the default way we experience stories, yet the ad experience still feels like a hard stop. We wanted to build a system where advertising respects the narrative and the viewer. Ads that feel native to the scene rather than a break from it. So, we set out to prove that ads can be contextual, subtle, and even helpful, without sacrificing the magic of the moment.
What it does
Seamless turns movies into shoppable moments without disrupting the story. We pre‑index every scene using VideoDB and an LLM layer to map transcripts to products. When playback starts, Dedalus Labs MCP invokes our shopping toolkit (Kroger + DoorDash) to fetch real inventory nearby. We then brand the in‑scene product with VideoPainter (open source model) overlay and surface it in a sidebar so viewers can checkout or save to cart instantly. The in-scene products are simultaneously stored into Snowflake DB for future savings on LLM calls.
How we built it
We built a preprocessing pipeline with VideoDB to ingest videos, create a scene index, and align transcripts with product candidates in scenes. On playback, an MCP server orchestrates our shopping agents to pull live product data and pricing.
The backend coordinates scene → product matching, returns overlay specs, saving to DB, while the frontend renders a Netflix‑style experience with a live product sidebar and shoppable actions. The demo showcases the full loop:
scene understanding → MCP fetch → overlay → checkout
Challenges we ran into
- Aligning scene context with product relevance in a way that feels natural.
- Bridging real‑time MCP tool calls with a smooth UI experience.
- Making overlays look native to the scene while preserving brand identity.
- Balancing technical depth with cinematic user experience
Accomplishments that we're proud of
- An end‑to‑end pipeline that moves from scene understanding to check out.
- Real inventory and pricing from shopping tools via MCP invoking.
- A polished Netflix‑style demo UI that makes the concept instantly graspable.
- A working overlay flow that shows how ads can live inside the story.
What we learned
We learned how powerful MCP orchestration can be for real‑time product discovery, and how essential strong preprocessing is for latency‑free personalization. We also learned that ad UX is just as important as recommendation quality, that subtlety and timing make or break the experience.
What's next for Seamless
- Personalization with richer viewer signals and preference learning.
- More retail partners and geographic coverage.
- Higher‑fidelity overlays and automated brand safety checks.
- Creator controls so placements respect narrative intent.
- A/B testing to optimize both experience quality and conversion.
- Clickable commerce for items already appearing in the scene.
- better decision making on when to inference w/ LLM versus pulling from DB
Log in or sign up for Devpost to join the conversation.