SSIA: Sustainability Strategy & Insights Agent

Sustainability teams want strategy briefs grounded in their carbon data. LLMs write fluent briefs but hallucinate figures, and in a domain headed for audit-grade assurance, one invented number destroys trust in the whole system.

Architect & product owner. Conceived and architected independently during my consulting tenure; it won a firm-wide AI & Sustainability innovation challenge. Rebuilt here as a clean-room demo.

• ▸One design principle above all: the math is deterministic and auditable, and the LLM never computes a number
• ▸A pure PCF engine computes cradle-to-gate footprints using process-based factors for hotspot components and input-output (EEIO-style) factors for the long tail, per the GHG Protocol Product Standard and ISO 14067
• ▸The agent writes a decarbonization brief over the computed rows only, with a citation on every claim; a second model judges factual consistency and citation coverage before anything is shown

• ▸Interactive PCF calculator: pick a product, configure the boundary, and see the footprint by lifecycle stage with per-row factor citations and an uncertainty range
• ▸Strategy brief with clickable citation chips and a Trust panel showing the judge's verdict
• ▸An architecture page mapping this lightweight demo 1:1 to a production AWS lakehouse design

• ▸Guardrails are demonstrated working, not described, including what happens when a claim fails verification
• ▸The calibration example reproduces its documented result exactly (9.4 kg CO₂e), because every factor is traceable