The International Standard for
Food Safety Data
SafeEats™ unifies government inspection records from 23 live data sources across 5 countries (US, UK, Canada, France, Netherlands), plus AI-assisted search for additional jurisdictions, into a single, normalized, commercially scalable dataset. Here's what makes it defensible.
23+
Live data sources
Direct government APIs
5
Live API countries
US, UK, Canada, France, Netherlands
7
Conversion archetypes
The normalization invention
1
Universal grade
A–F scale, where data exists
International Data Coverage
Three-tier strategy across live APIs and AI-assisted markets
Live API Markets
23 sources · 5 countriesDirect, real-time connectors to government open-data APIs — NYC, Chicago, SF, LA County, King County, Austin, Dallas County, Montgomery MD, Delaware, NY State, Boston, Houston, Toronto, UK FSA, France (Alim'confiance), Netherlands (NVWA), and more. Queried at request time, never stale snapshots.
- Socrata / CKAN / ArcGIS
- UK FSA + France + Netherlands national APIs
- Real-time freshness tracking
AI-Enhanced Markets
Additional jurisdictionsFor jurisdictions with publicly accessible inspection records but no structured API, AI reads official health department websites and portals. GPT-5 Mini identifies restaurants; Gemini 3 Flash pulls inspection data from official sources. Every result carries a confidence level and verification source — unverified entries are filtered out.
- Dubai Municipality (public sources)
- Denmark Smiley (findsmiley.dk)
- Confidence-labeled & deduplicated
Portal Redirect (Not Coverage)
Honest scopeFor jurisdictions with no scrapable data and no publicly accessible records, SafeEats links to the official health department portal so users can search manually. This is a convenience — not coverage. We do not claim the invention operates in jurisdictions where we can only redirect users elsewhere. The normalization requires data to normalize.
- Clearly labeled as redirect-only
- Not counted in coverage claims
- Honest scope protects data integrity
Inspection Grade Methodology
One universal A–F scale, auditable from source to score
Universal A–F Grade Scale
A
90–100
B
80–89
C
70–79
D
60–69
F
< 60
U
Unknown
Every jurisdiction's native scoring — penalty points, pass/fail, star ratings, compliance outcomes — is normalized to a single 0–100 scale. A score of 85 means the same thing whether the restaurant is in Los Angeles, London, or Toronto. Apples to apples, worldwide.
Normalization by Source
| Source | Native System | Conversion |
|---|---|---|
| NYC DOHMH | Letter grade A/B/C | Mapped directly; score from violation points |
| LA County DPH | Penalty points | Inverted: 100 − penalty points |
| King County, WA | Penalty points + result | Inverted; result shown on detail |
| Chicago CDPH | Pass / Fail + codes | Pass→high; Fail→F; weighted by severity |
| UK FSA (FHRS) | 0–5 star rating | Linear scale to 0–100 |
| France (Alim'confiance) | 4-tier evaluation | Result-based: 1→95, 2→82, 3→68, 4→40 |
| Netherlands (NVWA) | Compliance status | Voldoet→88; Niet voldoet→45 |
| AI-assisted | Public records | Gemini 3 Flash + confidence filtering |
Source provenance
Every record traces back to its issuing authority
Documented normalization
Conversion logic published — no black-box scoring
Relevance filtering
False matches stripped from permissive APIs
Unverified filtering
AI results with no verification are dropped entirely
Scalability of Restaurant Discovery
Architecture designed to scale from 23 sources to international coverage
Plug-in Data Connectors
Each government API is a self-contained connector with its own processor and normalizer. Adding a new market is a configuration task, not a rebuild — the pattern is proven across 23 live sources.
Unified Schema
One normalized record shape across every source. No per-jurisdiction parsing on the consumer side. A new data source slots into the existing search, detail, and history pipeline with zero frontend changes.
Tiered AI Pipeline
GPT-5 Mini for fast restaurant identification (2–4s), Gemini 3 Flash for real-time web search of official sources. Background-refresh pattern returns instant results, then silently upgrades with verified data.
Cache + Live Hybrid
Results are cached with freshness timestamps; live APIs are queried at request time where available. The system gracefully degrades — if an API is down, AI fallback kicks in automatically.
International by Default
The search engine handles any city, any country, any language. A user in Tokyo gets the same experience as a user in Austin — location-agnostic by design, not bolted on.
Low Marginal Cost
Adding the 24th, 50th, or 100th market follows the same proven pattern. The infrastructure scales horizontally — no new servers, no new frontend, no new schema. Just a new connector file.