Project goal: To see how AI can bridge the gap between digital convenience and physical enterprise spaces.

The main mission is to engineer a functional agent that solves a real-world challenge—specifically targeting industrial and commercial environments to eliminate energy inefficiencies.

Spatial Agent uses tools and capabilities to accomplish tasks (e.g., managing a local database, automating high-fidelity engineering workflows, and interacting with live database services).

My system can handle complex goals: the agent plans the steps and uses the tools at its disposal to finish the job.

Partner Power: Supported by the MongoDB for Startups program, my solution demonstrates a meaningful integration with MongoDB using MCP to give the agent its "superpowers" through high-performance data grounding.

For the build, I am using Google Cloud Agent Builder (rapid prototyping, building, and scaling).

Spatial Agent is a functional agent—powered by Gemini and Google Cloud Agent Builder—that integrates a MongoDB MCP server to solve a real industrial challenge.

Spatial Optician

💥 Inspiration

Traditional architectural evaluations are slow, disconnected, and lack real-time optical feedback. When designing spaces, architects have to manually calculate lux deficits, spatial efficiency, and lighting coefficients across static drawing boards and local databases. I wanted to bridge the digital convenience of generative AI with high-fidelity physical blueprint feedback. "Spatial Optician" was born to automate architectural visual analysis by letting an intelligent agent reason, plan, and analyze blueprints directly inside an immersive dashboard. I decided to leverage my core proprietary technology, Spatial Engine, and build a specialized B2B vertical—Spatial Optician—to solve the architectural lighting crisis.

📐 What it does

Spatial Optician is a complete visual calibration and spatial dynamics console.

Interactive Blueprint Workspace: A stunning, retro-futuristic glassmorphic blueprint interface featuring handwriting typography ("Architects Daughter") and pixel-perfect engineering grids.
Intelligent Spatial Agent: Powered by Gemini and Google Cloud Agent Builder, my agent performs multi-step missions: analyzes uploaded site photos, extracts depth buffers, and evaluates lighting anomalies.
Real-time Calibrations: Calculates optical scales, environmental factors (diffusion & Rayleigh scattering), and lux deficits.
MongoDB Grounding & Self-Healing Catalog (Partner Power): Supported by MongoDB for Startups, my agent interacts directly with a MongoDB cluster via MCP. If a requested light fixture is missing from the database, the agent automatically executes a live web search to find exact technical specifications, dynamically writes the new fixture back into the MongoDB catalog, and completes the ROI calculations on the fly.

🛠️ How I built it

Frontend: React, TypeScript, Vite, and Tailwind CSS v4 for the highly-interactive design. Powered by Spatial Engine core components for highly-interactive engineering grids and spatial visualization.
Animations: Framer Motion for smooth blueprint grids transitions.
Backend: FastAPI (Python) exposing spatial analysis and simulation endpoints.
Agent Core & Compliance: Built fully within the Google Cloud Agent Builder ecosystem using the official Google ADK (Agent Development Kit) in Python to coordinate the reasoning loops, sub-agents, and MCP tool execution.
Database & Integrations: MongoDB Atlas (powered by the MongoDB for Startups tier) and local instances, bridged via a custom TypeScript-based Model Context Protocol (MCP) Server running stdio transports.

🧠 Challenges I ran into

Environment Caching: Synchronizing path configurations and environment variables inside sandboxed shell runners during local setup. I bypassed this by utilizing uv for seamless Python workspace environment management.
Tailwind v4 Monorepo Config: Aligning path aliases and custom grid backgrounds in Tailwind v4 without traditional tailwind config files. I achieved this via raw css theme overlays (@theme) directly in index.css.
MCP Stdio Communication: Ensuring robust JSON-RPC communication between the Agent Builder and my local MongoDB instance through safe MCP tooling.

🏆 Accomplishments that I am proud of

Implementing a self-healing database feedback loop: building an agent that autonomously expands its own MongoDB product catalog when faced with missing technical specifications, using real-time search data and secure writebacks.
Designing an interface that wows at first glance, making static code look like a premium, active architectural terminal.
Fully automating the schema inspection, querying, and aggregate pipelines of MongoDB via a custom TypeScript MCP Server.
Zero manual environment setup requirement on the backend thanks to modern uv run packaging.

🎓 What I learned

I learned the immense power of the Model Context Protocol (MCP) in making LLMs "database-aware." Instead of writing brittle database API wrappers, I let Gemini autonomously reason about schemas, perform paginated queries, and inspect collections.

🔮 What's next for Spatial Optician

Integrate actual computer vision models on the FastAPI backend for real depth map extraction from images.
Deploy the MongoDB MCP server globally using secure Cloud Run instances with API keys authentication.
Expand the dashboard to support multi-layered CAD drawings rendering.