Carbon DT

Turn utility bills into actionable energy audits in seconds

Comment

1. Inspiration

Buildings account for a large portion of global energy consumption and carbon emissions, yet most operate without clear visibility into their energy performance. At the same time, many building owners are actively looking for ways to reduce utility costs. However, they lack the tools to understand where inefficiencies come from or what actions to take.

This gap is especially pronounced for small and mid-sized buildings, which are often excluded from traditional energy audits due to cost and complexity.

This challenge directly connects to UN SDG 7 (Affordable and Clean Energy), SDG 11 (Sustainable Cities and Communities), and SDG 13 (Climate Action).

We built Carbon DT to bridge this gap, enabling building owners to reduce both costs and carbon emissions through accessible, data-driven decision-making.

2. What it does

Carbon DT is a building energy audit platform that transforms basic utility and building data into actionable insights and reports.

From a technical perspective, the system:

  • ingests utility bills and building metadata
  • normalizes energy usage using weather data
  • predicts expected energy consumption
  • identifies inefficiencies through diagnostics
  • maps these to targeted recommendations
  • ranks them using a transparent scoring system
  • generates a structured PDF audit report summarizing insights, actions, and impact

From a social impact perspective, Carbon DT:

  • helps building owners reduce energy costs and financial waste (SDG 7)
  • improves operational efficiency and sustainability of buildings (SDG 11)
  • enables measurable reductions in carbon emissions (SDG 13)

By lowering the barrier to energy audits, Carbon DT makes sustainable decision-making accessible at scale, not just for large enterprises.

3. How we built it

We developed Carbon DT as a modular, explainable system designed to work with low-resolution data.

System architecture

1. Data ingestion & validation

  • Parses utility bill CSVs (electricity, gas)
  • Validates schema and normalizes units (kWh, therms)
  • Aggregates billing data into consistent monthly records

2. Feature engineering

  • Integrates weather data (HDD/CDD)
  • Computes energy intensity metrics (kWh/sqft, therms/sqft)
  • Extracts seasonal and trend features

3. Baseline modeling

  • Uses simple, interpretable models to estimate expected energy usage
  • Includes fallback heuristics when data is limited
  • Outputs predictions with confidence indicators

4. Diagnostics engine

  • Rule-based system that detects inefficiencies, such as high base load, seasonal overuse, and performance degradation over time
  • Each diagnostic includes severity and confidence

5. Recommendation system

  • Maps diagnostics to predefined expert recommendations
  • Includes attributes such as category, difficulty, and expected impact
  • Ranks recommendations using a weighted scoring framework that considers severity, confidence, implementation difficulty, and category weighting

6. Reporting layer

  • Streamlit-based UI for insight visualization and interaction
  • Generates downloadable PDF audit reports with:
    • building summary
    • energy performance insights
    • prioritized recommendations
    • cost and carbon impact summaries

Validation & Testing

We implemented validation across multiple levels to ensure reliability:

Unit testing

  • Verified unit normalization and aggregation logic
  • Tested emissions calculations and scoring outputs

Synthetic data testing

  • Created controlled datasets with known inefficiencies
  • Confirmed diagnostics correctly detect expected patterns

Baseline validation

  • Compared predicted vs actual energy usage
  • Evaluated residuals for consistency

End-to-end testing

  • Ran full pipeline using sample datasets
  • Ensured deterministic recommendation ranking and stable outputs

These tests were chosen to ensure correctness, consistency, and robustness under limited data conditions.

4. Challenges we ran into

Limited data availability Working with only utility bills (no BMS or sensor data) required careful feature engineering and fallback strategies.

Balancing accuracy and explainability We prioritized interpretable models and rule-based logic to maintain trust and usability.

Recommendation prioritization Designing a ranking system that reflects real-world decision-making — not just theoretical impact — was non-trivial.

From analysis to action Transforming insights into clear, prioritized recommendations required rethinking the system as a decision-support tool.

5. Accomplishments that we're proud of

  • Built a complete end-to-end energy audit MVP using minimal input data
  • Developed a diagnostic-driven recommendation engine that mirrors expert reasoning
  • Designed a transparent and explainable scoring system
  • Delivered actionable outputs including cost and carbon insights

What we learned:

  • Simple, well-designed systems can outperform complex models in low-data environments
  • Explainability is critical for user trust in decision-support tools
  • Prioritization (not just analysis) is key to real-world impact
  • A meaningful insight can be extracted from a data that may be overlooked

6. What’s next for Carbon DT

Near-term roadmap

  • Implement building-specific savings and carbon calculations
  • Improve recommendation precision and prioritization
  • Introduce user-specific optimization modes that let's the user choose the focus of the analysis, whether that be cost, carbon, or quick measures for an immediate result
  • Expand compliance-related features for real-world use cases

Long-term vision

As Carbon DT collects more building data, we plan to develop robust machine learning models to enhance prediction accuracy and personalization.

Ultimately, Carbon DT will evolve into a scalable energy intelligence platform. It will be enabling buildings everywhere to make smarter, lower-cost, and lower-carbon decisions, accelerating progress toward SDG 7, 11, and 13, and a better future for all of us.

Built With

Share this project:

Updates