Can AI Check Vastu From a Floor Plan? What It Reads, What You Confirm, and What It Cannot See
An AI can read the rooms off a floor plan, but it cannot know which way north is, and it does not invent the verdict. Which part is AI, and which is arithmetic.
Author - Vaastu AI Editorial
Yes — partly, and the part that is actually "AI" is smaller and more boring than the marketing around it suggests. Reading a floor plan is a vision problem: finding the walls, separating the rooms, picking up the door openings and the printed labels. That is what the model does. It cannot work out which way is north from a drawing, because that information is usually not in the image at all — so you confirm north and the room labels yourself. After that nothing is being imagined: each room's zone is looked up in a fixed direction-by-room table and weighted into a score. The verdict is arithmetic. What no AI can do is see anything the drawing does not contain.
The three stages, and only one of them is AI
| Stage | Who does it | What it decides |
|---|---|---|
| Read the plan | The model | Where the walls and rooms are, where the doors sit, what each room is probably called |
| Confirm | You | Which way is north, and whether "BED 2" is really a bedroom or the study you use it as |
| Score | A fixed table | Which zone each room falls in, the verdict for each room, and the overall number |
Most of the argument about whether AI can "do" Vastu collapses once those three are separated. Stage one is genuinely hard and genuinely automatable. Stage two is not automatable at all and every honest tool asks you for it. Stage three is a lookup — and a lookup is a good thing, because it means the same plan scores the same way twice.
Why north is the whole ball game
Get north wrong by one sector — 45 degrees — and every room in the plan shifts one zone. Nothing else in the pipeline has that kind of leverage. Here is the kitchen row of the table our check scores against, ranked:
| Kitchen sits in | Score | Verdict |
|---|---|---|
| South-East | 100 | Ideal |
| North-West | 80 | Good |
| South | 60 | Acceptable |
| East | 55 | Acceptable |
| West | 55 | Acceptable |
| South-West | 45 | Avoid |
| North | 35 | Avoid |
| North-East | 15 | Avoid |
The bands are fixed: 90 and above reads ideal, 70–89 good, 50–69 acceptable, below 50 avoid. So a kitchen that is genuinely in the south-east reads ideal at 100 — but rotate the plan one sector the wrong way and the same kitchen is reported in the east at 55, acceptable, or in the south at 60. Rotate it the other way and it lands in the north-east at 15, which reads avoid and would send you looking for remedies you do not need.
That is why a tool asking you to confirm north is not being lazy. It is the one input it cannot derive and the one that decides everything downstream. If you are not sure of yours, how to find the north direction of your house plan walks through the compass, the map and the architect's arrow, and where each of them goes wrong.
What is on a floor plan, and what is not
| The plan contains | The plan does not contain |
|---|---|
| Where each room sits relative to the others | Which way the building faces |
| Which wall the main door is on | The slope of the plot or the road level |
| Room shapes and rough proportions | What is above or below the room |
| The geometric centre of the unit | Materials, water sources, trees, neighbouring structures |
| The labels someone typed on it | Whether the drawing still matches the built flat |
Everything in the right-hand column is out of scope for any tool reading a file — ours included. A plan is a drawing of an outline, not a survey of a site. That is not a flaw to be fixed with a better model; the information simply is not in the input.
An AI that writes, and a table that looks up
There are two quite different things being sold as "AI Vastu", and it is worth knowing which one you are using.
The first is a general chatbot. You describe your flat in words, it writes back paragraphs. It never saw a plan, it is working from your description, and — this is the part that matters — ask it the same question twice and you can get two different answers, because generating text is not the same as computing one.
The second reads the plan and then hands the judgement to a published table. The model's job ends at "this room is here"; the verdict comes from the same lookup every time, for everyone. That is less impressive-sounding and considerably more useful, because you can audit it. If a table says a north-west toilet scores 90 and reads ideal — and on the bathroom row, north-west is the only placement that reaches the ideal band at all — you can disagree with the table on the merits. You cannot disagree with a paragraph that changes each time you ask.
What our free check actually does
It detects the rooms on the plan you upload, stops and asks you to confirm north and the room labels, then shows which zone each room falls in and a compliance score built from that table. Rooms are not weighted equally in the score — the main entrance carries 18%, the kitchen 16%, the master bedroom 13%, and each toilet 12% — so the number tells you what to look at first. The paid detailed report adds the room-by-room analysis and remedies sorted by how much of the deficit they actually recover. You can run the free check on your own floor plan, and if you want to see the zone grid the scoring sits on before you upload anything, the Vastu grid tool draws it over a plan for you.
Four questions worth asking any AI Vastu tool
- Does it ask you to confirm north, and let you correct it? If it does not, ask how it decided.
- Does it show you which zone it put each room in? A score with no zone map cannot be checked.
- Does it publish the table behind the verdict? Without it, the number is an opinion with a decimal point.
- Does it distinguish what it read from the plan from what it assumed? Anything about slope, surroundings or what sits above a room was assumed — it is not in your file.
An AI can read your floor plan. It cannot see your site, and it should not be the one deciding what the reading means.