Executive one-pager · WattIQ EPM Decision Agent

An AI that turns model outputs into grounded, cited, validated decisions

A planner asks in plain language; the system reads the World Bank–ESMAP EPM workbooks, reasons under a governed policy, cites every number to its source, and improves from use — without ever trusting an unverified figure. It now turns those results into scene-based galleries, theater-grade presentations, and decision-ready exports with a preserved narrative arc.

Runtime Grounded data Learning Platform

Ask

A planner's question, framed for their role — ministry, finance, regulator, operator.

Secure

Access gate + rate limit on the hosted service; proxy-resilient sign-in.

Reason & ground

The governed model (OpenAI gpt-5.5) calls data tools that read the EPM sheets, and cites every figure to workbook · sheet · row.

tool-call loop · cites [n]

Assemble

A role-specific brief and charts built server-side from the same figures.

Present

Decision cards, interactive charts, a Charting Station gallery, and a projected Theater deck in the browser.

One-click export to Word, PDF, PowerPoint, PNG, ZIP PNG galleries, or portable HTML — branded, scene-labeled, and decision-ending aware.

Learning loop

Observe every Q&A→ Store in Azure→ Recall as advice→ Validate & promote

gate: ≥2 agree + human approval — the model proposes, a human disposes

Ships & runs

GitHub Docker Render Azure Blob store · git-tag rollback

Why it can be trusted

Provenance or nothing — every number re-derived from the source sheets and cited each run. Governed by trust tiers — the AI can suggest, never rewrite its own rules or safety. Human-validated learning — nothing is trusted until corroborated and approved.

WattIQ EPM Decision Agent · reasoning model: OpenAI gpt-5.5Runtime: Python 3.11 Docker · APIs: OpenAI-compatible + Azure BlobDemo Designed and Developed by Jonathan Davidar · © 2026 Jonathan Davidar. All Rights Reserved.