What shipped
Built a runnable Python 3.11+ local audit kit for Local AI Visibility Radar. The CLI can create a demo audit, generate query packs and manual capture templates, ingest captured result JSON, score visibility and source quality, and render client-ready Markdown reports.
Main commands:
``bash PYTHONPATH=src python -m local_ai_visibility_radar demo --out audits/nashville-demo PYTHONPATH=src python -m local_ai_visibility_radar build audits/nashville-demo bash scripts/smoke.sh ``
Architecture
- Standard-library Python CLI via
argparseto keep setup simple. - JSON files for business profiles and captured results so an operator can inspect and edit data by hand.
- Deterministic query generation, scoring, action prioritization, and Markdown rendering with no required API keys.
- Separated modules for models, query generation, scoring, reports, and demo fixtures.
- Reports are rendered as plain Markdown for easy client delivery, editing, or later conversion to PDF.
Trimmed scope
- No SaaS app, accounts, database, billing, or hosted UI.
- No scraping or automated assistant/browser capture.
- No external LLM summarization, even optionally, in this build.
- No PDF export; Markdown is the delivery artifact for the MVP.
- No multi-location batch workflow.
Limitations
- Manual result capture is required for real audits.
- Fact matching is intentionally simple and transparent; abbreviations such as
AveversusAvenuemay be flagged for operator review. - Scoring is directional and should be calibrated with more paid audit examples.
- Surface volatility, personalization, and geography can change results between captures.
- Demo data is fixture-based and must not be presented as live market evidence.
Suggested next steps
- Add a small Streamlit review UI for filling
raw_results.json. - Add optional PDF export from the Markdown client report.
- Add a calibration file for category-specific scoring weights.
- Add screenshot attachment indexing.
- Add an optional OpenAI-assisted summary mode guarded by
OPENAI_API_KEY, while keeping deterministic mode as the default. - Add monthly monitoring comparisons between two audit workspaces.