Selected work / Case study
AI Interior
Design Agent
Archisketch · Graphics Engineer → R&D Lead · Apr 2025 – Jul 2026
01 / Brief → Floor plan
Start with a room.
The walkthrough begins with a simple request: designing a studio apartment. The assistant explains the available starting points, and the user chooses to draw a room directly in the editor.
A roughly 18 m² plan becomes the shared context for the conversation. The user can move between direct manipulation of the plan and instructions in natural language.
02 / Style request → Layout alternatives
Compare possibilities in 3D.
A request for a warm, cozy atmosphere produces three furnished proposals. Each option includes a preview and an estimated cost, so the user can compare both spatial character and budget.
The proposals remain editable 3D scenes. The walkthrough switches between plan and perspective views, compares the alternatives, and confirms one before continuing with furniture edits.
03 / Selected design → Itemized quotation
Connect the layout to its cost.
After selecting a proposal, the user asks how much it will cost. The agent opens an itemized quotation containing the furniture, dimensions, quantities, and prices associated with the current design.
A follow-up message requests the quotation as a PDF. Layout decisions and their cost can be reviewed from the same conversation, without starting a separate estimating workflow.
04 / Auto Camera → Rendered views
Ask for a view of the space.
The user then asks for camera views that cover the studio. The agent generates five viewpoints and returns previews in the conversation.
The user requests renders of all five views. The resulting images can be opened and inspected alongside the editable model, connecting spatial arrangement with the way the room is presented.
05 / Multi-agent systems / Spatial graphics
Engineering a connected workflow.
My engineering work covered the multi-agent Auto Layout and Auto Camera pipelines, translating natural-language requests into furniture arrangements and camera configurations.
I also worked on the spatial data structure used by the agents and implemented optimization algorithms for 3D model compression and content streaming. These systems supported the browser-based design experience shown above.
06 / Role / Release / Demonstration
From development to delivery.
I joined Archisketch as a Graphics Engineer in April 2025, became R&D Lead in November 2025, and led AI Interior Design Agent development as project lead and PM until July 2026.
We shipped the agent to paid subscribers, connecting floor-plan analysis, 3D conversion, style recommendation, furniture layout, rendering, and cost estimation through natural-language input.
I represented Archisketch at MWC and Amazon Unicorn Day through product demonstrations, and presented the end-to-end workflow built on Amazon Bedrock AgentCore at Amazon Unicorn Day (AWS Startups).