Skip to content
Lab preview · competitive intelligence

EnergyRadar

A local-first cockpit for tracking Portuguese energy retail offers with evidence attached to every change.

BuildingPlaywright · SQLite · FastAPI · Streamlit
What we’re trying to solve

Competitive intelligence with receipts.

Energy offers change often, but a report without the original page, screenshot, and timestamp is difficult to trust. EnergyRadar explores a repeatable way to monitor the market while keeping the underlying evidence visible.

System

Capture first. Interpret second.

Playwright captures competitor pages and stores raw HTML, cleaned text, screenshots, and source URLs. Deterministic extraction and optional local-model analysis then produce offer records, change events, and reports.

High-level workflow

source pages → browser capture → evidence store
                         ↘ offer extraction → diff → review → report
Evidence

What the repository makes visible.

EvidenceWhat the repository shows
CapturePlaywright-based capture stores raw HTML, cleaned text, screenshots, and source URLs.
AnalysisOffer and change records reference the snapshot and screenshot that support them.
ResilienceOllama is preferred, with deterministic regex extraction when it is unavailable.
OperationsSeed, capture, extract, diff, report, daily, and run commands are documented.
What we’re learning

Freshness is not the same as truth.

01 · Evidence outlives summaries

A concise insight is useful only when the underlying page and capture remain inspectable.

02 · Determinism earns trust

Fallback extraction provides a stable baseline for comparing model-assisted results.

03 · Source maintenance is product work

Selectors, page changes, and coverage need explicit ownership before this can become a dependable service.

04 · Local-first is practical here

SQLite and filesystem storage keep an evidence prototype easy to inspect and operate.

Next

What comes next.

The next milestone is a dated, reproducible evaluation of source coverage, extraction accuracy, and change-review effort across a stable test set.