EarthOps — AI Command Center

Autonomous AI sustainability command center powered by Gemini 3.7 Flash that detects facility anomalies, quantifies CO2e ROI, and generates certified decarbonization work orders.

Comment

Inspiration

Commercial and institutional buildings account for 39% of global carbon emissions, yet physical facility operators report that up to 30% of energy and water is wasted due to silent mechanical faults—such as stuck chiller bypass valves, simultaneous heating and cooling, night-time HVAC schedule drift, and undetected baseline leaks.

Most sustainability platforms are merely passive dashboards that deliver retroactive ESG compliance reports months after utility bills arrive. We built EarthOps to bridge the gap between massive environmental data streams and real-time physical action, transforming raw multi-sensor telemetry into autonomous root-cause forensics, EPA-validated emissions accounting, and dispatchable engineering work orders.

What it does

EarthOps is an Autonomous AI Sustainability Command Center powered by Gemini 3.7 Flash and deterministic thermodynamic physics engines:

  • Multi-Sensor Telemetry Normalization: Ingests high-frequency interval readings for Electricity (kWh), Water (kL), Waste Diversion (kg), Ambient Outdoor Temp (°C), and Badge Occupancy.
  • Autonomous Forensic Agent (Gemini 3.7 Flash): Analyzes anomalous off-hours consumption, cross-correlating building physics with occupancy to identify specific mechanical root causes (e.g., Air Handling Unit economizer damper stuck open at 40°F).
  • Emissions & Financial ROI Quantification: Translates physical waste into Scope 1 & Scope 2 greenhouse gas footprints ($MT\text{ CO}_2e$) using official EPA eGRID 2024 regional factors and computes direct utility savings ($/year).
  • Multi-Criteria Action Prioritization: Ranks facility engineering interventions balancing carbon impact, financial payback, and implementation feasibility.
  • CMMS Maintenance Dispatch: Generates certified maintenance work orders complete with equipment tags, mechanical room locations, and OSHA Lockout/Tagout (LOTO) safety protocols.
  • Interactive What-If Decarbonization Simulator: Allows sustainability directors to model rooftop solar PV additions, night HVAC setbacks, graywater recycling, and composting programs against SBTi Net-Zero 2030 trajectories.
  • Zero-Hallucination Guardrails & Copilot: Clearly tags data provenance ([OBSERVED DATA] vs [AI INFERENCES]) with an integrated conversational Copilot for on-demand operational briefings.

How we built it

  • AI & LLM Reasoning: Google Gemini 3.7 Flash via the official @google/genai TypeScript SDK for multi-step reasoning, anomaly diagnostics, and executive report synthesis.
  • Frontend: React 18, TypeScript, Tailwind CSS, Lucide Icons, and Motion for reactive animations.
  • Data Visualization: Recharts (Composed Charts, Area baselines, Bar distributions) and D3.js for interactive time-series telemetry.
  • Backend Architecture: Node.js & Express API with full-stack TypeScript compilation (tsx & esbuild).
  • Standards & Compliance: EPA eGRID 2024 emission factors, GHG Protocol Corporate Accounting Standard, ISO 50001 Energy Management, and Science Based Targets initiative (SBTi).

Challenges we faced

  • Cross-Correlating Heterogeneous Telemetry: Reconciling disparate measurement frequencies (15-minute electrical pulse meters vs. daily municipal water and weekly waste logs) into a unified time-series baseline.
  • Eliminating AI Hallucinations in Engineering: Ensuring Gemini outputs realistic equipment failure modes grounded in actual sensor numbers rather than generalized advice. We solved this with a dual-engine architecture: deterministic thermodynamic baselining provides boundary constraints, while Gemini handles causal diagnosis.
  • High-Density Industrial UX: Designing an information-dense yet scannable command center that operates seamlessly across desktop monitors, tablets, and mobile screens without UI clipping.

Accomplishments that we're proud of

  • Sub-Second Forensic Pipeline: Built an end-to-end multi-agent pipeline that ingests 45 days of multi-building sensor data, identifies anomalies, and generates prioritized engineering actions in under 2 seconds.
  • Actionable Physics over Passive Metrics: Translating abstract numbers into tangible mechanical solutions (e.g., repairing valve actuator #3 with a 0.3-year payback).
  • Interactive Decarbonization Sandbox: Enabling facility managers to test capital expenditure scenarios dynamically with instant visual recalculation of ESG scores and emissions curves.

What we learned

  • Multi-agent LLM systems provide immense value when combined with domain-specific mathematical guardrails (thermodynamics + EPA conversion factors).
  • Facility operators care most about actionable feasibility, safety compliance (LOTO), and short payback windows rather than vanity metrics.

What's next for EarthOps

  • Automated BACnet / Modbus IoT Ingestion: Direct plug-and-play integrations with existing Building Automation Systems (Siemens Desigo, Honeywell Forge, Johnson Controls Metasys).
  • Automated Carbon Credit Verification: Instant export of ISO 14064 verifiable decarbonization credits for voluntary carbon markets.
  • Predictive Grid Demand Response: Automated peak shaving using battery storage and chiller pre-cooling when municipal electricity prices spike.

Built With

Share this project:

Updates