The Techem EcoCoach (Prediction Engine) is an intelligent, predictive Digital Energy Platform designed to transition real estate portfolios from retrospective utility billing to real-time, AI-driven climate neutrality optimization.
By leveraging massive IoT edge data, historical consumption, and external APIs (like live weather), the platform forecasts three critical dimensions:
- Energy Consumption (kWh)
- CO₂ Emissions (g/kWh)
- Heating Costs (€)
This platform provides actionable intelligence for both Tenants (to reduce daily consumption) and Landlords (to plan retrofits and optimize portfolio ROI), aiming to significantly reduce the CO₂ footprint generated by the residential sector.
The platform splits into two distinct, role-based experiences:
Empowers individual renters to monitor, predict, and optimize their energy usage.
- Today's Consumption & Budgeting: Real-time tracking of current daily usage and progress towards user-defined cost/CO₂ reduction targets.
- Predictive Forecasting: Time-series charts displaying historical actuals and future forecasts based on split-conformal prediction models.
- Room-Level Drilldown: Identifies which rooms are consuming the most energy and their heat-loss sensitivity.
- AI Conversational Coach: A Gemini-powered LLM chat interface providing personalized, data-backed energy-saving advice.
- Anomaly & Leak Detection: Identifies irregular consumption spikes indicating potential leaks or inefficiencies.
- Peer Comparison: Cohort benchmarking with leaderboards and gamified savings targets.
Tailored for property managers to oversee building portfolios and plan ESG compliance.
- Portfolio KPIs: High-level dashboard aggregating total MWh, costs, CO₂ tons, and average Energy Classes across all properties.
- AI Thermal Insights: Identifies systemic thermal issues (e.g., severe heat-loss units) across the portfolio using automated sensitivity analysis.
- Retrofit ROI Advisor: Scenario planner calculating the financial return, carbon tax savings, and expected property value increases for insulation or heating system upgrades.
- Automated ESG Reporting: AI-generated executive summaries for investors detailing portfolio performance and compliance trajectories.
The application is built with a modern, high-performance tech stack:
- Framework: React 18, Vite, TypeScript
- Routing: React Router v6
- State Management: React Query (
@tanstack/react-query) - Styling: Tailwind CSS,
shadcn/ui(Radix UI primitives)
- Framework: Python, FastAPI
- Machine Learning: PyTorch, XGBoost/LightGBM (for time-series forecasting and regression)
- LLM Integration: Google Gemini (for conversational agents and narrative generation)
- External APIs: Meteostat (for live dynamic weather forecasting)
The platform tackles high-scale IoT data with a 4-layer advanced machine learning stack (L0–L3):
graph TD
%% Define color classes for the nodes
classDef l3Node fill:#e1bee7,stroke:#8e24aa,stroke-width:2px,color:#000
classDef l2Node fill:#bbdefb,stroke:#1e88e5,stroke-width:2px,color:#000
classDef l1Node fill:#c8e6c9,stroke:#43a047,stroke-width:2px,color:#000
classDef l0Node fill:#ffccbc,stroke:#e53935,stroke-width:2px,color:#000
%% L3
subgraph L3 ["L3: Application & AI Agents (Business Logic)"]
UI("Tenant & Landlord Dashboards"):::l3Node
LLM("Gemini LLM (Conversational Coach)"):::l3Node
UI --- LLM
end
%% L2
subgraph L2 ["L2: Quantile & Counterfactual Models"]
Conformal("Conformal Calibration (10th/90th Bounds)"):::l2Node
WhatIf("Counterfactual Analysis (Thermal Sensitivity)"):::l2Node
end
%% L1
subgraph L1 ["L1: Baseline Machine Learning"]
XGB("XGBoost / LightGBM Time-Series"):::l1Node
Feat("Feature Engineering (Weather, History)"):::l1Node
Feat --> XGB
end
%% L0
subgraph L0 ["L0: Hierarchical Reconciliation"]
Raw("Raw IoT Sensor Data (Unit Level)"):::l0Node
Recon("Reconciliation (MinT / Room Shares)"):::l0Node
Raw --> Recon
end
%% Flow (Connecting specific nodes to ensure 8.8.0 compatibility)
Recon -->|Disaggregated Data| Feat
XGB -->|Point Forecasts| Conformal
Conformal -->|Calibrated Predictions| UI
WhatIf -->|Calibrated Predictions| UI
%% Add background colors and dashed borders to the subgraphs themselves
style L3 fill:#f3e5f5,stroke:#ab47bc,stroke-dasharray: 5 5,stroke-width:2px
style L2 fill:#e3f2fd,stroke:#42a5f5,stroke-dasharray: 5 5,stroke-width:2px
style L1 fill:#e8f5e9,stroke:#66bb6a,stroke-dasharray: 5 5,stroke-width:2px
style L0 fill:#fbe9e7,stroke:#ff7043,stroke-dasharray: 5 5,stroke-width:2px
- L0 - Hierarchical Reconciliation (MinT): Disaggregates unit-level consumption down to individual rooms.
- L1 - Baseline Machine Learning: Core forecasting using advanced tree-based models and exogenous features.
- L2 - Probabilistic Forecasting & What-If: Uses quantile regression to provide calibrated uncertainty bounds (10th/90th quantiles) and evaluates behavioral changes (e.g., "What if I lower the heat by 1°C?") using learned room-level thermal sensitivities.
- L3 - Online Application & AI: The presentation and conversational layer. Tenant and landlord LLM features degrade gracefully to tabular data if the API becomes unavailable (Offline Resilience).
The repository consists of a frontend and a backend application.
- Navigate to the
srcdirectory (or wherever your backend app resides). - Install the required dependencies.
- Set your environment variables (including
GEMINI_API_KEY). - Run the FastAPI server:
uvicorn techem.serve.api:app --app-dir ../src --port 8123 --reload
- View the interactive Swagger API documentation at
http://localhost:8123/docs.
-
Navigate to the
frontenddirectory. -
Install dependencies:
npm install
-
Start the Vite development server:
npm run dev:frontend
Note: Currently, the frontend is partially driven by a robust mock data layer (
src/lib/mockData.ts). Full backend integration involves swapping these mock objects with live React Query hooks pointing to the FastAPI endpoints.
- Carbon Tax Predictor: Project future CO₂ emission costs based on impending regulatory hikes.
- Solar PV Potential Estimator: Automatically calculate rooftop solar yield.
- Predictive Maintenance: Detect hardware faults in boilers or individual radiators before a breakdown occurs.