CarbonPilot

AI Builder Cup 2026 · Sustainability & Social Impact

Companies can measure their carbon. Few can decide what to do about it.

CarbonPilot is an AI carbon decision engine for companies in Asia-Pacific. It turns fragmented operational data into ranked, costed, explainable actions. Gemini plans the analysis; a deterministic engine produces every number.

Walk the demo company →

The problem we answer

Companies can measure their carbon footprint, but turning fragmented operational data into prioritised, financially meaningful decarbonisation actions remains difficult.

“Build an AI-powered solution that addresses a meaningful environmental or societal challenge.”[1] The theme asks for tools that improve how resources are used and help people make better-informed, sustainable decisions. A footprint number alone does neither.

Who has the problem

The sustainability or operations manager who owns the number and has to act on it. They ask four questions, and each screen answers one:

  1. 01 Where are our biggest emissions, and how sure are we? Measure →
  2. 02 Why did they move this month? Diagnose →
  3. 03 What can we cut, and what will it cost? Simulate →
  4. 04 Can I justify it to management? Act →

The decision gap, in one company’s data

Live from the engine
  1. 1

    Raw data

    168 activity records

    14 activity streams across 5 source files (procurement.csv, electricity.csv, fuel.csv, travel.csv, cloud_usage.csv), 12 months.

  2. 2

    The footprint

    11,208 tCO₂e / year ± 3,225

    With each emission factor’s published uncertainty propagated through. This is where a carbon report stops.

  3. 3

    The choices

    4,294,967,296 possible plans

    Every on/off combination of the 32 actions the engine found, before deciding how far to push each one. Carbon, cost, effort and confidence pull in different directions.

  4. 4

    The decision

    −33.8% carbon, saves $775K/yr

    Asked to cut at least 15% at no extra operating cost, the engine searched 2,098 combinations; 342 qualify. The best one, in 4 actions: Electrify the delivery fleet, Renewable PPA for purchased electricity, Mandate rail over short-haul flights, Supplier engagement and low-carbon sourcing.

    Run the search yourself →

PT Nusantara Digital is a synthetic 12-month dataset of an Indonesian digital-services company, built for this demo with anomalies planted for the detector to find. Upload your own CSV on the Data page to run the same loop on real numbers.

Why it matters here

Asia-Pacific is where the decision has to be made

The region emits most of the world’s carbon, makes most of the world’s goods, and is now putting deadlines and prices on both. Most of its companies have no team to turn a footprint into a plan.

60%[2]

of the world’s greenhouse-gas emissions come from Asia-Pacific

60.1% in 2024, up from 41.7% in 1991

26×[3]

a company’s supply-chain emissions versus its own operations, on average

Scope 3 vs Scope 1+2, companies disclosing to CDP in 2023

97%+[4]

of ASEAN businesses are micro, small or medium enterprises

~70 million firms, 85% of employment; few have a sustainability team

Same kilowatt-hour, very different carbon

Lifecycle carbon intensity of 20 JAPAC power grids, gCO₂e per kWh, cleanest first[5]

Indonesia ranks 19 of 20: 705 g/kWh, 69% coal, 7.6× New Zealand. That is why purchased electricity is 28.5% of this company’s emissions on 1.5% of its spend, and why its single largest cut is renewable PPA for purchased electricity (−2,452 t/yr).

Indonesia (home grid) Other JAPAC grids
  • New Zealand 93
  • Pakistan 308
  • Sri Lanka 368
  • South Korea 417
  • Vietnam 464
  • Japan 477
  • Singapore 497
  • Cambodia 499
  • Myanmar 502
  • Australia 525
  • China 528
  • Thailand 546
  • Philippines 595
  • Malaysia 606
  • Taiwan 634
  • India 671
  • Hong Kong 676
  • Bangladesh 696
  • Indonesia 705
  • Mongolia 822

Ember intensities are full-lifecycle (generation plus plant construction, fuel supply chains and upstream methane), not the generation-boundary factor used for Scope 2 reporting. The company’s reported Scope 2 uses the national inventory factor, so this chart compares grids and never enters the inventory. Source: Ember Yearly Electricity Data, CC BY 4.0.

Deadlines and prices are arriving

Disclosure mandates and carbon prices across the region

  1. Sep 2023

    Indonesia Carbon price

    IDXCarbon opens; carbon trading becomes mandatory for coal-fired power plants.[6]

  2. Jan 2025

    Australia Disclosure

    Mandatory climate reporting (AASB S2) starts for the largest entities; mid-size from Jul 2026, the rest from Jul 2027.[7]

  3. Jan 2025

    Hong Kong Disclosure

    All listed issuers must disclose Scope 1 and Scope 2 emissions.[8]

  4. FY2025

    Singapore Disclosure

    All listed companies report Scope 1 and 2; Straits Times Index companies add Scope 3 from FY2026.[9]

  5. 2025

    Malaysia Disclosure

    Large Main Market issuers report under the ISSB-based framework; the rest of the market follows in 2026–27.[8]

  6. Jan 2026

    Singapore Carbon price

    Carbon tax rises from S$25 to S$45 per tonne, heading for S$50–80 by 2030.[10]

  7. Jan 2026

    EU → JAPAC exporters Carbon price

    CBAM enters its definitive phase: steel, aluminium, cement, fertiliser, electricity and hydrogen sold into the EU carry a carbon cost.[11]

  8. 2026

    Malaysia Carbon price

    Carbon tax introduced in stages, starting with iron, steel and energy.[12]

  9. FY to Mar 2027

    Japan Disclosure

    SSBJ sustainability standards become mandatory for Prime-market companies of ¥3tn+; ¥1–3tn follow a year later.[13]

Each one puts a deadline or a price on a footprint. A report meets the deadline. Choosing the right actions is what protects the margin.

Why we built it this way

Measure. Diagnose. Simulate. Act.

Existing platforms help companies measure and report carbon. CarbonPilot is built for the moment after: which action to take next, why, and what it will do to carbon and to cost. Each step runs a named method on the data, so every answer can be checked.

  1. 01Measure

    How much, and how sure?

    Activity × emission factor under the GHG Protocol, stream by stream, each factor with its source, year and ± uncertainty. Uncertainty is propagated to the total, not hidden.

    14 factors · ±29% Open →

  2. 02Diagnose

    What moved, and why?

    A robust z-score (median and MAD) on every stream’s monthly series, so a spike cannot inflate its own baseline and hide; then attribution to the activities and usage patterns behind it.

    4 anomalies in 12 months Open →

  3. 03Simulate

    What if we change it?

    A deterministic scenario engine and an exhaustive constraint search: a carbon target and a budget in, every feasible combination of interventions out.

    2,098 plans searched Open →

  4. 04Act

    What do we do first?

    Every action scored on carbon, cost, effort and confidence, read as a marginal abatement cost curve, re-ranked under a carbon price, then assigned an owner.

    32 actions ranked Open →

Gen AI where it earns its place

Gemini plans. The engine computes.

Ask a question in plain language. Gemini on Vertex AI decides which analyses to run, calls the engine’s tools, cites the methodology it relied on, and explains the result with its assumptions. It also reads CSV columns the rules cannot map. It never writes a carbon number itself: every figure in an answer came back from a tool call, and the answer shows each call.

13 engine tools the agent can call

  • get_baseline

    Get the company's total carbon footprint in tCO2e, total cost, and the Scope 1/2/3 split.

  • get_emission_breakdown

    Break total emissions down by category, activity_type, or scope.

  • get_monthly_series

    Month-by-month emissions and cost, optionally filtered to one activity type or category.

  • compare_month

    Compare a YYYY-MM month against the preceding six months, with per-activity-type deltas.

  • detect_anomalies

    Find statistical anomalies in the monthly emission series using a robust median/MAD z-score.

  • explain_anomaly

    Explain WHY an anomaly happened: baseline comparison plus attributed drivers (e.g. weekend share shift).

  • run_scenario

    Simulate what-if interventions and return estimated carbon, financial impact and assumptions.

  • find_feasible_scenarios

    Constraint search. Given a minimum carbon reduction and a maximum cost increase, enumerate intervention combinations and return which ones satisfy both constraints.

  • search_emission_factors

    Search the emission factor library for a source, unit, scope and uncertainty figure.

  • search_methodology

    Retrieve the carbon-accounting methodology passages behind a claim: GHG Protocol scope definitions, Scope 2 location vs market based, spend-based Scope 3 caveats, grid and conversion-factor sources, Cloud Carbon Footprint method, uncertainty propagation, and robust spike detection. Call this whenever you explain HOW a figure was derived or which scope/method applies, then cite the returned source and section.

  • lookup_live_factor

    Corroborate an emission factor against Climatiq's live published library, and optionally estimate kgCO2e from activity data with it. Call this when asked whether the factors are real, current, or sourced from a commercial provider. The curated factor library remains the source of truth for this company's reported numbers.

  • list_recommendations

    List ranked decarbonisation actions with carbon impact, cost impact, effort and confidence.

  • create_action

    Create a tracked action item from an approved recommendation.

Sources

  1. [1] AI Builder Cup 2026: Challenges & Requirements, Sustainability & Social Impact
  2. [2] UN ESCAP, Review of Climate Ambition in Asia and the Pacific 2025
  3. [3] CDP & BCG, Scope 3 Upstream: Big Challenges, Simple Remedies (2024)
  4. [4] ASEAN Secretariat, Development of MSMEs in ASEAN
  5. [5] Ember, Yearly Electricity Data (CC BY 4.0), via the CarbonPilot engine
  6. [6] ANTARA, Indonesia launches carbon exchange IDXCarbon (Sep 2023)
  7. [7] ASIC, Sustainability reporting
  8. [8] Latham & Watkins, ISSB Implementation Across Asia (Sep 2024)
  9. [9] XBRL International, Singapore eases climate reporting timelines (2025)
  10. [10] National Environment Agency Singapore, Carbon tax
  11. [11] European Commission, Carbon Border Adjustment Mechanism
  12. [12] MIDA, Budget 2025: carbon tax on iron, steel and energy by 2026
  13. [13] Japan FSA, roadmap for SSBJ-based disclosure (Apr 2026)
  14. [14] GHG Protocol Corporate Standard

Company figures on this page are computed live by the CarbonPilot engine. Regional figures are published sources, reviewed 16 October 2026.