Computational Engine

AI where it counts.
Rigor everywhere else.

The Koi Engine builds avoided emissions models on demand, turning solution descriptions into customized, auditable forecasts backed by years of methodology development and a curated climate data lake. It produces in seconds what used to take weeks of manual analysis.

How It Works

Four stages. Full transparency at every one.

The engine does not replace scientific judgment. It accelerates the structured, repeatable parts so scientists can focus on the decisions that actually require expertise.

Stage 01

Identify

Maps each climate solution to the baseline it displaces and the value chains it touches. No forecast is credible without knowing what you are comparing against.

Draws from curated datasets spanning IEA, SBTi, US Federal LCA Commons, and peer-reviewed literature to select the right reference points.

Stage 02

Match

Aligns identified value chains to GHG intensity and market data in the Koi Data Lake. Enforces dimensional analysis so intensity multiplied by market size produces meaningful avoided emissions.

Automatic unit conversion across energy (Wh, J, BTU), mass, distance, and economic value. Incompatible units are caught, not papered over.

Stage 03

Fill

Where the Data Lake lacks a specific data point, AI synthesizes relevant literature to fill the gap. Every AI-sourced value is flagged on the model Datasheet.

Market capture estimation uses SEC filings and sales data to approximate real deployment timelines. This is the first automated method that avoids blanket market uptake assumptions.

Stage 04

Produce

Outputs a fully customizable, version-controlled model with transparent inputs, documented assumptions, and audit-grade provenance.

Every model is evergreen. When underlying Data Lake information changes, models update automatically, keeping forecasts consistent across technologies and time.

Grounded in established data

IEASBTiUS Federal LCA CommonsSEC FilingsPeer-Reviewed Literature

Scaffolding & Guardrails

AI accelerates. Methodology governs. The engine is where they meet.

Close to a decade of climate modeling expertise defines every constraint, every validation, and every default the engine operates within. AI accelerates the execution. It does not make the decisions.

Pre-Validated Data Lake

Curated sources (IEA, SBTi, US Federal LCA Commons, academic literature) form the foundation. The engine reasons on top of verified data, not raw web scrapes.

Unit Compatibility Enforcement

Dimensional analysis is enforced at every stage. If emissions per mass meets a market denominated in value, the engine resolves or rejects. It never silently proceeds.

AI Data Flagging

Every value sourced by AI is explicitly marked on model Datasheets. Scientists and users know exactly which data points were synthesized and which came from the Lake.

Plausibility Bounding

Automated models are bounded for plausibility before release. Three QA tiers (Rapid, Refined, Diligence) layer increasing human review on top of engine output.

Auditability

Every input traceable. Every assumption documented.

Speed means nothing if you cannot defend the output. The engine is built so that every forecast can withstand the scrutiny of investors, auditors, and regulators.

Version Control

Every model versioned. Select prior or future versions with full change tracking.

Transparent References

Data sources, boundaries, and assumptions documented. Nothing is hidden behind a black box.

Edit Mode

Review, modify, and customize any model while maintaining the audit trail. Your changes, your reasons, all on the record.

QA Datasheets

Tags indicate manual scientist review level, data verification status, and AI-sourced values. Everything visible at a glance.

Executive SummarySolution ScaleUnit ImpactDatasheet

Viewing Version

1.1 Latest

Access Level

Owner

QA Tier

Refined

Last Updated

2026-01-21

Forecast Configuration

Annual results between 2026 and 2036

Addressable Market

milk production (1,100 Mt milk)

Market Capture

1% in 10 years (Start: 2025)

Baseline

Milk

Solution

Fermentation protein-based dairy alternative (thin consistency)

References

Current configuration references

Baseline Scenario
Clune S, Crossin E, Verghese K, Systematic review of greenhouse gas emissions for different fresh food categories, Journal of Cleaner Production (2016). Science-Based Targets initiative (SBTi). Sector Guidance, 2024.
Solution Scenario
Smetana, S., Mathys, A., Knoch, A., & Heinz, V. (2015). Meat alternatives: life cycle assessment of most known meat substitutes. The International Journal of Life Cycle Assessment, 20(9), 1254-1267.
Addressable Market
https://openknowledge.fao.org/server/api/core/bitstreams/60012716-5b7d-418a-81b9-543bf6245ca4/content

Illustrative datasheet. Actual models contain full reference chains and methodology notes.

Years of methodology.
Seconds to a forecast.

See the engine in action on a solution you care about. Every input, every assumption, every output. No hand-waving.