MOVE 03
Complex data, made simple
The data ran deep, but reading it took a forecaster.
Units, colours and model names assumed expertise most users don’t have.
01
Layers grouped by domain
Shipped
A maritime user scrolled past eight atmospheric layers to reach wave height. Now split into Atmosphere and Marine.
02
Right panel rebuilt around the location readout
Shipped
Account controls led and the reading was buried. Login moved to the menu; the selected place’s conditions took the top.
03
Search as the primary way in
Proposed
Users panned and zoomed to find sites they visit daily. I proposed search-first with auto-suggest. It missed this cycle, and it’s still the biggest open problem on the map.
04
Suggestions from recent searches
Shipped
Where a user looked before is the best guess at where they’ll look next. It turns typing into a tap.
05
Saved locations
Shipped
The team’s sites, each with its alert state, so the morning check happens before the map opens.
06
Familiar positions stay where they are
Shipped
Layers stay on the left, the timeline at the bottom and the model picker at the bottom right. Changes roll out in steps, and no control is moved and restyled in the same release.
Keeping it readable as it grows
Overlays and buttons had equal weight, and the map lost. Then the platform grew from about 20 features to 90, and each new one needed a place.
BEFORE
Fifteen pills in one stack.
01
AFTER
Grouped by domain, in one list.
TWO MORE DECISIONS
Interface states, and one altitude control for pressure levels.
Show decisions 07–08
BEFORE
Eight level chips, inside the Wind row.
ALTITUDE
500
hPa
5,574 m
FL183
1000
850
700
500
300
200
08
AFTER
One slider, with metres and flight level alongside.
07
Interface states: loading, selected, empty, stale
“No data” and “still loading” looked identical. For an operator deciding whether to halt a crane, that difference is the whole decision.
08
Altitude control on pressure levels
hPa with flight levels shown alongside, so aviation users never convert in their heads.
09
The phone screen, from original to redesign
The same map screen at three points.
ORIGINAL
Where it started
CURRENT APP
Where it is today
REDESIGN
Where it is headed
02
The detailed forecast, with a graph behind each row
Same grid and row order as the live table. Each row now carries a line, bar or colour band, so the trend reads before the numbers.
BEFORE
The live table on weatherex.ai: numbers in a grid, read cell by cell.
AFTER
The Basic panel, Kochi, 3-hourly.
A
B
C
D
A
Each day leads with its high and its rain
↑ 30° · 2.9 mm sits on the day row, apart from the hours below.
B
Feels like shares the temperature chart
A dashed line over the actual. The shaded gap shows how much hotter it feels.
C
Rain as bars, sized by amount
Height grows by square root, so light rain still shows next to heavy rain.
D
Gusts in wind-speed colours
Calm stays clear. Colour climbs from cyan to red, so strong spells jump out.
THE KEY BEHIND IT
One colour rule per variable, shared by Basic, Waves & Tides, Meteogram and Ocean.
Show the colour key
11
Atmosphere
7
Marine · Waves & Tides
6
Ocean
5
Chart rules
24
variables
3
groups
4
forecast views
1
colour rule each
MOVE 04
From data to decisions
The platform showed readings and left the judgement to the user. This move is about a product that says what the weather means for your day or your site.
Inference cards and the two AI dashboards, shown below, were accepted in principle and wait on model capability, so they are not shipped.
01
The assistant given the weight it was designed for
Shipped
Cumulus looked like a footnote on a map it was meant to lead. Its entry point now matches its role.
02
Simple by default, Pro for depth
Prototype
General users want a verdict. Operational users need the raw values to make their own call. Simple mode opens first, and Pro keeps the model and layer depth.
03
Alerts that read before the text
Prototype
Each alert carries a sparkline, a severity frame, a condition glyph and a time-to-impact bar. Severity and timing land before the words do.
04
Model comparison, side by side
Prototype
A split-screen map and comparative dashboards, so users can see where the models disagree.
Inference cards: the platform reads the location for you
Prototype
Today it shows layers, values and model runs. No inference, no alerts. Users interpret it themselves, every day.
Cards that read a location and return a reading with an action attached. Not "40–60 mm rain overnight" but "cover moisture-sensitive cargo by Thu 20:00", with the numbers one tap away.
SAVED PLACE
Matheran
Maharashtra
Monsoon in full swing
This weekend · 5–6 Jul
Go for the mist,
not the views.
It stays wet both days — this is Matheran at its greenest and most dramatic, but the famous viewpoints will likely be fogged in. Lovely if you’re after waterfalls and quiet forest trails.
Great for
waterfalls, greenery, fewer crowds
Tough for
viewpoints, horse rides, dry feet
High confidence
HOW WE KNOW
24°
Feels like 27°
Weekend rain
70–90 mm
Rainy days
Both
Humidity
94%
Best window
Sat PM
UPDATED 09:04 IST
ECMWF
Two AI dashboards: split by job
Prototype
One forecast, two jobs. A general user wants to know whether to go out. A planner needs the numbers, because the thresholds live in their SOPs. Same data; the split is in interpretation.
1
2
3
4
FIG. 04 — PROPOSED · GENERAL USER · MOBILE
1
Verdict up front
The answer is the headline. Nobody should read a map to decide when to leave.
2
One confidence read
Five models collapse into a single meter. No model names, no grid sizes.
3
The dry window comes first
Rain bars exist to show one thing: when it is dry.
4
Numbers on demand
Temperature, rain and wind sit one tap away, collapsed by default.
MOVE 03
Complex data, made simple
The data ran deep, but reading it took a forecaster.
Units, colours and model names assumed expertise most users don’t have.
01
Layers grouped by domain
Shipped
A maritime user scrolled past eight atmospheric layers to reach wave height. Now split into Atmosphere and Marine.
02
Right panel rebuilt around the location readout
Shipped
Account controls led and the reading was buried. Login moved to the menu; the selected place’s conditions took the top.
03
Search as the primary way in
Proposed
Users panned and zoomed to find sites they visit daily. I proposed search-first with auto-suggest. It missed this cycle, and it’s still the biggest open problem on the map.
04
Suggestions from recent searches
Shipped
Where a user looked before is the best guess at where they’ll look next. It turns typing into a tap.
05
Saved locations
Shipped
The team’s sites, each with its alert state, so the morning check happens before the map opens.
06
Familiar positions stay where they are
Shipped
Layers stay on the left, the timeline at the bottom and the model picker at the bottom right. Changes roll out in steps, and no control is moved and restyled in the same release.
Keeping it readable as it grows
Overlays and buttons had equal weight, and the map lost. Then the platform grew from about 20 features to 90, and each new one needed a place.
BEFORE
Fifteen pills in one stack.
01
AFTER
Grouped by domain, in one list.
TWO MORE DECISIONS
Interface states, and one altitude control for pressure levels.
ALTITUDE
500
hPa
5,574 m
FL183
1000
850
700
500
300
200
BEFORE
Eight level chips, inside the Wind row.
08
AFTER
One slider, with metres and flight level alongside.
07
Interface states: loading, selected, empty, stale
“No data” and “still loading” looked identical. For an operator deciding whether to halt a crane, that difference is the whole decision.
08
Altitude control on pressure levels
hPa with flight levels shown alongside, so aviation users never convert in their heads.
09
The phone screen, from original to redesign
The same map screen at three points.
ORIGINAL
Where it started
CURRENT APP
Where it is today
REDESIGN
Where it is headed
02
The detailed forecast, with a graph behind each row
Same grid and row order as the live table. Each row now carries a line, bar or colour band, so the trend reads before the numbers.
BEFORE
The live table on weatherex.ai: numbers in a grid, read cell by cell.
AFTER
The Basic panel, Kochi, 3-hourly.
A
B
C
D
A
Each day leads with its high and its rain
↑ 30° · 2.9 mm sits on the day row, apart from the hours below.
B
Feels like shares the temperature chart
A dashed line over the actual. The shaded gap shows how much hotter it feels.
C
Rain as bars, sized by amount
Height grows by square root, so light rain still shows next to heavy rain.
D
Gusts in wind-speed colours
Calm stays clear. Colour climbs from cyan to red, so strong spells jump out.
THE KEY BEHIND IT
One colour rule per variable, shared by Basic, Waves & Tides, Meteogram and Ocean.
11
Atmosphere
7
Marine · Waves & Tides
6
Ocean
5
Chart rules
24
variables
3
groups
4
forecast views
1
colour rule each
MOVE 04
From data to decisions
The platform showed readings and left the judgement to the user. This move is about a product that says what the weather means for your day or your site.
Inference cards and the two AI dashboards, shown below, were accepted in principle and wait on model capability, so they are not shipped.
01
The assistant given the weight it was designed for
Shipped
Cumulus looked like a footnote on a map it was meant to lead. Its entry point now matches its role.
02
Simple by default, Pro for depth
Prototype
General users want a verdict. Operational users need the raw values to make their own call. Simple mode opens first, and Pro keeps the model and layer depth.
03
Alerts that read before the text
Prototype
Each alert carries a sparkline, a severity frame, a condition glyph and a time-to-impact bar. Severity and timing land before the words do.
04
Model comparison, side by side
Prototype
A split-screen map and comparative dashboards, so users can see where the models disagree.
Inference cards: the platform reads the location for you
Prototype
Today it shows layers, values and model runs. No inference, no alerts. Users interpret it themselves, every day.
Cards that read a location and return a reading with an action attached. Not "40–60 mm rain overnight" but "cover moisture-sensitive cargo by Thu 20:00", with the numbers one tap away.
SAVED PLACE
Matheran
Maharashtra
Monsoon in full swing
This weekend · 5–6 Jul
Go for the mist,
not the views.
It stays wet both days — this is Matheran at its greenest and most dramatic, but the famous viewpoints will likely be fogged in. Lovely if you’re after waterfalls and quiet forest trails.
Great for
waterfalls, greenery, fewer crowds
Tough for
viewpoints, horse rides, dry feet
High confidence
HOW WE KNOW
24°
Feels like 27°
Weekend rain
70–90 mm
Rainy days
Both
Humidity
94%
Best window
Sat PM
UPDATED 09:04 IST
ECMWF
Two AI dashboards: split by job
Prototype
One forecast, two jobs. A general user wants to know whether to go out. A planner needs the numbers, because the thresholds live in their SOPs. Same data; the split is in interpretation.
1
2
3
4
FIG. 04 — PROPOSED · GENERAL USER · MOBILE
1
Verdict up front
The answer is the headline. Nobody should read a map to decide when to leave.
2
One confidence read
Five models collapse into a single meter. No model names, no grid sizes.
3
The dry window comes first
Rain bars exist to show one thing: when it is dry.
4
Numbers on demand
Temperature, rain and wind sit one tap away, collapsed by default.