Energy Copilot

energy.soufien.lu — Luxembourg Energy Transition Co-Pilot

Live product · Architecture · Solution assessment · Flagship decision

An AI-assisted prospecting workspace for Luxembourg solar-PV opportunities. It combines roofs, car parks, tenders, market data, deterministic tariff calculations, and cited Luxembourg/EU law in a map-first workflow.

What this demonstrates

• A C#/.NET 10 agent host using Microsoft Agent Framework abstractions and application-owned

safety, stopping, validation, and deterministic tool boundaries.

• Natural-language-to-SQL over PostgreSQL/PostGIS with a read-only allowlist and bounded results.
• Python data pipelines for public geospatial data, tenders, document extraction, and vision.
• Azure AI Content Understanding for tender predicates and multimodal classification for existing PV.
• React, Material UI, and MapLibre for a spatial workflow that can export the filtered evidence.
• Terraform and GitHub Actions deployment to Azure, with telemetry and explicit evaluation evidence.

Legal text is not copied into this repository. The bounded law tool consumes the public Lex MCP server, which owns the Luxembourg/EU temporal corpus, provenance, search indexes, and ingestion lifecycle.

Architecture

React/MapLibre workspace → .NET agent API → bounded tools
                                           ├─ PostGIS data finder / NL→SQL
                                           ├─ deterministic tariff engine
                                           ├─ tender finder
                                           ├─ energy-document RAG in pgvector
                                           └─ law.soufien.lu MCP client

public datasets → Python pipelines/jobs → PostgreSQL/PostGIS/pgvector

The architecture case study records current state, decisions, governance, risks, and the deliberate gap to an enterprise deployment. Binding technical details live in CLAUDE.md and docs/copilot-technical-spec.md. The separate solution assessment frames value, feasibility, process redesign, cost drivers, controls, and pilot exit gates without claiming unmeasured ROI.

Repository layout

• api/ — .NET API, orchestrator, tools, and deterministic services.
• web/app/ — React/MapLibre product UI.
• pipelines/ — Python ingestion, enrichment, and data-quality jobs.
• eval/ and evals/ — deterministic and live-model evaluation evidence.
• docs/ — architecture, ADRs, runbooks, and dataset provenance.
• fixtures/ and probes/ — publisher/API samples and contract checks.

Run and verify

The complete cloud product depends on Azure services and public datasets; it is published primarily as a reproducible case study rather than a zero-infrastructure demo.

dotnet build -c Release
dotnet test

Set-Location web/app
npm ci
npm run build

Set-Location ../../pipelines
python -m pytest

Deterministic tests run on ordinary changes. Live-model evaluations are deliberate pre-release runs and must not be described as if every commit incurs their cost.

Security and limitations

• Public-data product; no user accounts or long-term user profiles.
• Model output cannot directly mutate the database or invent tariff numbers.
• Legal answers are citations from Lex, not legal advice.
• The public deployment is a personal-scale architecture; the case study names the controls an

enterprise rollout would add rather than claiming they already exist.

License

Apache-2.0. See [LICENSE](LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: SFHAJJI
• Source: SFHAJJI/energy-copilot
• License: Apache-2.0
• Homepage: https://energy.soufien.lu

Install and usage instructions live in the source repository linked above.