1
00:00:00,000 --> 00:00:05,371
Sunshift helps a person plan flexible household
electricity use around expected solar

2
00:00:05,371 --> 00:00:10,742
energy. Its opening pairs a calm planning
workspace with an imagined solar

3
00:00:10,742 --> 00:00:16,114
neighborhood, created with AI image generation.
The illustration is not a photographed

4
00:00:16,114 --> 00:00:21,485
installation. This is a narrated interface
walkthrough using actual app screenshots and

5
00:00:21,485 --> 00:00:22,380
synthetic narration.

6
00:00:23,583 --> 00:00:27,916
The working planner brings an hourly energy
chart and appliance timing into

7
00:00:27,916 --> 00:00:32,248
one workspace. The starting day is an editable
sample, not live meter

8
00:00:32,248 --> 00:00:36,581
data. This view has the Sunshift plan selected,
with six point seven

9
00:00:36,581 --> 00:00:40,913
five kilowatt hours of modeled grid import. A
person can explore which

10
00:00:40,913 --> 00:00:45,246
loads might use more of the solar energy they
expect, then compare

11
00:00:45,246 --> 00:00:47,773
the same inputs with the original schedule.

12
00:00:49,000 --> 00:00:53,522
The original view shows the starting schedule
for this illustrative sunny day.

13
00:00:53,522 --> 00:00:58,043
It calculates twenty two point nine five
kilowatt hours of grid import.

14
00:00:58,043 --> 00:01:02,565
The sample Sunshift plan shown earlier
calculates six point seven five. That

15
00:01:02,565 --> 00:01:07,087
difference comes from shifting load within the
same model and assumptions. It

16
00:01:07,087 --> 00:01:11,609
is not a claim about a measured household
saving. The useful question

17
00:01:11,609 --> 00:01:15,000
is which input and timing choices explain that
difference.

18
00:01:16,208 --> 00:01:21,346
The energy inputs are deliberately visible.
There are twenty four hourly solar

19
00:01:21,346 --> 00:01:26,484
estimates and twenty four household consumption
estimates. Each value represents kilowatt hours

20
00:01:26,484 --> 00:01:31,622
in that hour. A sunny sample, a cloudy sample,
or an all

21
00:01:31,622 --> 00:01:36,760
zero profile can help explore a different day.
Household consumption excludes the

22
00:01:36,760 --> 00:01:41,898
flexible appliances being scheduled, so the same
demand is not counted twice.

23
00:01:41,898 --> 00:01:44,038
These remain user supplied assumptions.

24
00:01:45,250 --> 00:01:50,050
Select an appliance to edit its power, whole
hour duration, and allowed

25
00:01:50,050 --> 00:01:54,850
running window. The entire uninterrupted run
must fit between the earliest start

26
00:01:54,850 --> 00:01:59,650
and the finish by time. An original start
defines the comparison schedule.

27
00:01:59,650 --> 00:02:04,450
A fixed start can preserve a personal
constraint. Changing an input invalidates

28
00:02:04,450 --> 00:02:09,250
the old plan and asks the user to optimize
again, keeping old

29
00:02:09,250 --> 00:02:10,850
calculations from appearing current.

30
00:02:12,083 --> 00:02:16,828
The optimize action calculates a schedule within
those windows and locks. Original

31
00:02:16,828 --> 00:02:21,572
and Sunshift plan views compare the same day
using different appliance start

32
00:02:21,572 --> 00:02:26,317
times. Grid import is computed hour by hour
whenever demand exceeds solar

33
00:02:26,317 --> 00:02:31,061
generation. Solar self consumption shows the
share of generated solar used in

34
00:02:31,061 --> 00:02:35,806
that modeled day. Any displayed improvement is a
calculation for these inputs,

35
00:02:35,806 --> 00:02:38,573
not a measured reduction from a household.

36
00:02:39,792 --> 00:02:44,442
A named scenario can be saved in this browser
and reopened later.

37
00:02:44,442 --> 00:02:49,093
The scenario library supports multiple days and
variants, so an experiment with

38
00:02:49,093 --> 00:02:53,744
different assumptions does not have to erase the
previous idea. The saved

39
00:02:53,744 --> 00:02:58,395
charge at noon example shown here is a separate
illustrative variant. Scenario

40
00:02:58,395 --> 00:03:03,046
JSON provides a portable backup, and a schedule
can be exported as

41
00:03:03,046 --> 00:03:07,696
a CSV or printed. There is no account or cloud
sync; clearing

42
00:03:07,696 --> 00:03:10,022
browser data removes locally saved scenarios.

43
00:03:11,250 --> 00:03:16,471
The method page explains how the model works.
The scheduler tries alternative

44
00:03:16,471 --> 00:03:21,692
starts and improves them through a bounded
heuristic. It is deterministic, but

45
00:03:21,692 --> 00:03:26,913
it does not guarantee a globally optimal
combination. Overlapping appliance loads are

46
00:03:26,913 --> 00:03:32,134
allowed and reported. There is no battery,
tariff, weather feed, carbon factor,

47
00:03:32,134 --> 00:03:37,355
or circuit capacity model. Those limits matter
before anyone translates an exploratory

48
00:03:37,355 --> 00:03:39,530
schedule into real appliance use.

49
00:03:40,750 --> 00:03:46,248
Sunshift's code, design, documentation, and
tests were developed with OpenAI Codex
assistance.

50
00:03:46,248 --> 00:03:51,746
No AI service runs inside the product:
scheduling is implemented in TypeScript.

51
00:03:51,746 --> 00:03:57,244
The result is a focused planning loop from
editable assumptions to a

52
00:03:57,244 --> 00:04:02,742
calculated schedule and a reusable export. Its
purpose is to make timing

53
00:04:02,742 --> 00:04:08,240
choices understandable, while keeping the
evidence, the calculation, and the limits
visible.