TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Studio Practice · AGE 18—21 · FACTUAL · DURABLE
Site research with a fast assistant
Take the model's questions, and answer each one from a primary source.
When to use
The first fortnight of any project — site, context, climate, programme — when the question list is long and you do not yet know what you do not know.
The method
Four moves. One: describe the site and the brief to a model and ask for the questions a thorough site analysis would answer — the questions, not the answers. Two: sort them — keep, merge, drop, and add the ones it missed, which will be the local ones. Three: answer each kept question from a primary source — the site visit, the survey, the municipal record, the meteorological data, the neighbour — and write the source beside the answer. A resident who has watched the lane flood every monsoon for thirty years is evidence, like the drainage layer; that a dataset is easier for a machine to read does not make it the stronger source. Four: build the dossier with two visible layers: what the AI suggested, and what a real source confirmed. Anything the model answered on its own stays marked as a lead until a source arrives.
Watch for this
The model answering the questions it was asked only to pose. It will, fluently, with the climate of a city three hundred kilometres away. Take the questions; discard the answers.
Try it
Pick a real plot you can visit. Get a model's question list. Walk the site with the list and answer what you can by looking. Count how many questions only the visit could answer, and how many the model had quietly answered wrong before you went.
Prove it
Show one site dossier with a source cited for every fact the design leans on, and one AI-drafted question that genuinely changed where you looked.
How it works
Models are strong at the shape of an enquiry — what a site analysis usually covers — because they have read thousands of them. They are weak at your site, because it was never in the data. Systems with search attached can fetch public documents, which shortens step three for some facts. It does not replace the visit, the survey, or the conversation with the person next door. This method was first taught in the Lab's own workshops and is kept there as a module.
Site research is also the first of several places where AI enters a studio without making any images. The same discipline — draft fast, then verify against a source — runs through scheduling and programme checks, specification drafting, the small scripts and code a studio needs, and the early-stage simulation surrogates that estimate daylight or heating and cooling loads before a full model exists. Each of those has its own card on this map with its own verification rules. The habit you are learning here is the one they all share.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.