I’m a product engineer shipping spatial and business intelligence end-to-end, across model orchestration, computer vision, and generative interfaces.

Ask a building questions about it in natural language you would to a Project Manager to cut onboarding time.

Talk to a Building

(

Text-to-SQL

/

Gemini 2.5 Flash

/

iTwin.js

)

Talk to a Building

(

Text-to-SQL

/

Gemini 2.5 Flash

/

iTwin.js

)

A floor plan image becomes queryable spatial data, with room and furniture count and types.

Floor Plan Raster to Vector

(

SAM 2

/

GeoJSON

/

React

)

Floor Plan Raster to Vector

(

SAM 2

/

GeoJSON

/

React

)

Agentic loop to generate, critique, and refine masks until guardrail requirements are met.

Agentic Image Validation Loop

(

Python

/

OpenCV

/

Google ADK

)

Agentic Image Validation Loop

(

Python

/

OpenCV

/

Google ADK

)

Natural language turns into costed, live-updating office programming dashboards.

Brief to Spatial Dashboard

(

Next.js

/

Vercel AI SDK

/

Tailwind CSS

)

Brief to Spatial Dashboard

(

Next.js

/

Vercel AI SDK

/

Tailwind CSS

)

Logic behind how a grounded LLM turns language into structured numbers and metrics tailored to interior designers.

Adaptive Interface Logic

(

Gemini 2.5 Flash

/

Vercel AI SDK

/

Prisma

)

Adaptive Interface Logic

(

Gemini 2.5 Flash

/

Vercel AI SDK

/

Prisma

)

Turning photos spatial for quick visualizations on an infinite canvas for designers.

Spatial Photos on Canvas

(

tldraw SDK

/

Gaussian Splatting

/

WebGL

)

Spatial Photos on Canvas

(

tldraw SDK

/

Gaussian Splatting

/

WebGL

)

Software is learning to see. Models are reasoning about space, and making decisions on real-world data, and industries are being rebuilt around them.

I’ve been building from inside that shift: an infinite canvas platform orchestrating 30+ generative models; LLM features turning natural language into structured spatial programs for ~1800 designers; and an agent pipeline that reads floorplans and returns structured geometry.