The engine behind SnowSure
How the Concierge knows
what it knows.
The SnowSure Concierge is powered by the Snowdata Answer Engine — a grounded AI built around seven weather models, decades of historical archive, and verified ground truth from snow safety teams at the world's ski resorts. Here's how it works, what it can do, and why every answer is something we stand behind.
The principle
The AI is the writer. The data is the source.
The single most important thing to know about the Snowdata Answer Engine: it doesn't make things up.
Every question runs through a grounded pipeline. The engine assembles a verified evidence pack at the moment you ask — drawing from live resort conditions, the seven-model forecast consensus, decades of historical pattern matching, and any operator-approved FAQs the resort has provided. It then composes a clear, factual response, citing the evidence where helpful.
If the data isn't there, the engine tells you so. We'd rather give you an honest “we don't have that yet” than a confident-sounding guess.
Snowfall numbers, base depths, scores, and forecasts always come from verified Snowdata sources at query time. They never come from the model's weights. Periodic fine-tunes sharpen the engine's voice and phrasing — never the facts.
How it works
Four steps. Most of them invisible.
01
Parse the question
Identify the language, the intent, and the resort scope. Route to global, resort-specific, or comparative paths.
02
Assemble the evidence
Pull live conditions, intelligence cards, season history, forecast model outputs, operator FAQs, and any promoted Q&A that matches the question.
03
Compose the answer
The language model writes a concise, factual response based on the evidence. Surfaces uncertainty where the data warrants it.
04
Log and learn
The question, the evidence, the answer, the latency, and any feedback are logged. Resort operators can approve corrections, which feed the next answer.
The data underneath
Built on the world's best snow inputs.
SnowSure doesn't run its own atmospheric model. We combine the world's best:
7 weather models reconciled
NOAA / GFS · ECMWF · Météo-France ARPEGE · Met Office UKV · JMA GSM · BOM ACCESS · MetService NZ. Reconciled per resort, every hour.
30+ years of archive
Every model's forecast measured against verified snowfall at every resort. This is what makes per-resort forecast skill measurable.
50+ live data sources
SNOTEL and international equivalents (CR2, NIWA, BOM, MeteoSchweiz), satellite snow cover (NASA MODIS, ESA Sentinel), webcam vision analysis, ski-patrol attestations.
SnowSure ML · Beta
The learned layer.
SnowSure ML is the model we built to weight the seven numerical weather predictions against verified outcomes — SNOTEL stations, resort attestations, and historical forecast verification. Currently shadow-run for days 8–14 of the forecast horizon. Published as a separate signal alongside the seven-model consensus — never blended into the average. Labeled Beta until forecast skill demonstrably exceeds the consensus over a full season.
How it learns
Like a great ops team. Not like AI mythology.
Every question the Concierge answers is logged. Resort partners can see the most frequent visitor questions in their admin dashboard, approve strong answers, and add official FAQs in five languages. Improvements apply on the next ask — not after a model retrain.
Periodic fine-tunes from operator-approved Q&A sharpen the engine's voice and phrasing over time. The facts — the numbers, the forecasts, the scores — always come from verified Snowdata at the moment you ask.
It's a grounded knowledge loop, not “AI training itself.” A real ops team is the closest analogy.
Log everything
Every question writes to the answer-engine queries log: question, resort, intent, evidence, answer, latency, session.
Feedback
Thumbs up / down and optional corrections flow into a moderation queue.
Operator FAQs
Resort admins author official FAQs in EN, ES, FR, DE, IT. Injected into the evidence pack on matching questions.
Promoted corrections
Operator-approved corrections are pulled in as learned evidence on future matching questions.
Analytics
Frequent questions surface in the admin dashboard. Resorts see what to FAQ next.
Voice fine-tune
Approved Q&A feeds periodic fine-tunes — sharpens phrasing only. Snowfall numbers never come from the model's weights.
Where the engine lives
One engine. Many surfaces.
The Snowdata Answer Engine powers:
SnowSure Concierge
Here on snowsure.ai. The consumer-facing surface — designed for travelers, in editorial voice.
→ /conciergeSnowdata Answer Engine
On snowdata.ai. The B2B and developer surface — same engine, technical register.
→ snowdata.ai/answer-engineAI assistants worldwide
Claude, ChatGPT, Grok, Gemini, Perplexity, Copilot — and any MCP-aware agent. Travelers asking AI about snow are reaching the same engine, with full attribution.
Resort white-label embed
The same engine, branded for the resort, on their own website or lodge displays. Visitor questions logged for partner analytics.
REST API + MCP
For developers integrating directly. Free read tier for individuals; commercial tier for production use.
→ /developersBuilt for the open agentic web
Discoverable. Verifiable. Yours to connect to.
SnowSure is among the first ski and snow services to be conformance-tested on Google's Agentic Resource Discovery (ARD) standard. We publish a cryptographically signed catalog of our capabilities (ai-catalog.json) at a well-known path on our domain, so any ARD-compatible AI agent can discover us, verify our identity, and connect.
We're also listed in the official MCP Registry as ai.snowsure/snow. Compatible with Copilot Agent Finder, Hugging Face hf discover, and any standards-aware client.
Use it. Build on it. Tell us how it could be better.
Try the Concierge
The fastest way to understand the engine: ask it something.
Build with it
MCP server, REST API, JS embed for resort sites. Free read tier. Five languages.
Developer docs →Verify our claims
Read our methodology, see the live signed catalog, check the MCP Registry listing.
Methodology →
