New Yorkers filed 300,000+ no-heat complaints last winter.
HeatAhead reads the weather forecast and ranks every residential building by its
chance of losing heat before the cold snap arrives, so inspectors can knock
before tenants freeze.
Same risk, history and owner record the map shows. Building only: never type an apartment number.
69.1%of the top 100 failed within the exact 48 hours (best simple rule: 61.7%)
96.6%of the top 100 lost heat within 30 days of being flagged
6 of 6past winters where the model beat the 14-day recency baseline
481,356/sbuildings scored per second with Snap ML on an IBM z17 LinuxONE guest
Start here
Not sure where to start?
Answer a few quick questions and HeatAhead lays out the steps that fit you,
whether your apartment is cold tonight or you want a building ready before the first
freeze.
Under a minuteNo sign-in neededNothing is saved unless you choose
Your plan
Ask the data
Have a specific question?
Ask about NYC heat complaints in plain words. HeatAhead pulls the public
data, draws the charts and works out the numbers for you.
An hourly job on the LinuxONE
instance reads the Open-Meteo 7-day forecast and looks for two or more freezing days in a row.
2
Decide
The model re-scores every building under
the forecast temperatures (the whole city in about a third of a second on the s390x guest) and
ranks them for inspection.
3
Respond
People who saved a building over the
calibrated threshold get one email for that spell; inspectors get the ranked queue.
Get HeatAhead updates
Hear about new features and what we're working on. These are project updates, separate from your building's heat alerts.
No heat right now?
Don't wait for a score or an alert. Tell your landlord, then report it to NYC311 with your
name and apartment number so HPD can inspect.
Maps and photos use external providers when you view them. These settings control HeatAhead's optional browser storage.
About external services
Ask HeatAhead
Help with the site
Not sure how to word it? Try a few words. We'll suggest close matches.
Please leave out passwords, email addresses, phone numbers and apartment numbers.
No close matches. Try fewer words or choose All topics.
Getting started
Map and risk
Data and privacy
Accounts and alerts
What to do
How HeatAhead works
1
Check the weather
Forecasts show when freezing weather is expected.
2
Look at past problems
Past complaints, violations and building records help the model estimate heating risk.
3
Choose what to do
Tenants and owners can review a building. Inspectors can see which places the model suggests checking first.
A student project by NYU Tingineers for the IBM Z Datathon. The model work ran on IBM LinuxONE. HeatAhead is independent of NYC311 and HPD.
Read the map
Color shows winter risk
Darker blue means higher estimated risk. A hexagon shows the average score of the properties inside it. Zoom in to see each property's colored dot.
Height shows past complaints
Taller hexagons mean more heating complaints were recorded last season. They don't show the number of buildings or their actual height.
Click a hexagon
You'll see a list of buildings, with higher scores first. Use Next to see more. Click a building in the list to open its record.
Zoom in for a closer look
The city view uses larger hexagons so it's easier to read. Zoom in to see smaller areas, then colored dots for each property. Gray building shapes come from the background map. One property record can cover more than one building.
Winter risk covers the whole season. Cold-day risk looks at a shorter period around freezing weather. They cover different lengths of time, so the numbers can be different.
See how it was checked
On a building's page, open How accurate is this number? It compares predictions with what happened in a winter the model wasn't trained on. This helps check the model overall, but can't say for sure what will happen to one building.
A high score doesn't mean the heat has failed. A low score doesn't rule out a problem.
Some heating problems never get reported, so they may be missing from the records.
From 20% up, scores are shown as ranges. No score goes above 99%.
If a building has no score, we don't have an estimate for it. It can still have heating problems.
Check your building's history and the available owner details. Sign in to save a building for email alerts. Emails also need the team's alert service to be running.
We use public NYC records about buildings, heat complaints, violations, owners and boilers, along with weather data. Some records may be missing or out of date. A complaint means someone reported a problem; it doesn't mean an inspector confirmed it.
The map is about buildings, not individual tenants. Saving a building for alerts links it to your verified email. HeatAhead doesn't see your Google password. Owner and agent details come from public business records.
When you sign in with Google, we receive only your verified email address, your name and Google's account id for you; we keep no Google access tokens and ask for no Drive, Gmail or contacts access. An email sign-in link uses your address the same way. Your saved buildings, alert choices and dashboard panels are stored on the team's own server and are never sold or shared. You can remove a saved building or stop every alert from your profile at any time, and signing out clears the sign-in cookie on that device.
The help chat uses published answers. It can suggest close matches when a question is unclear or has a typo. Common questions stay on your device. If AI matching is enabled, an unfamiliar question and the previous help topic may be sent to a provider to suggest a published question. Your Google account details and earlier messages aren't included. We don't save guide chats in this app. Please leave out passwords, contact details and apartment numbers.
The photos were taken earlier; they aren't live. The guide can show you where to go. Sending a message, reporting a problem or changing alerts is still up to you.