சென்னை / a working record
The Madras Project.
After Madras Day I got stuck on a simple question: were we celebrating a city that began in 1639, or a British foothold on a coast that was already a set of older places? I am now building the coast the question is about.
Many places. One Chennai.
The log.
How this started and how it is going, one leaf at a time, with the renders and the numbers, including the failures. Turn the leaves.
10 notesTen things that surprised me.
Sourced claims about Madras before it was Chennai, each one I am happy to lose. Where the question started.
340 nodesMethod.
What is evidence, what is illustrative, how each frame is judged, and every source the world resolves to.
about $2,700Ledger.
What it has taken: sessions, hours, tokens, renders, disk, and a rough cost at list prices. Estimates, recalibrated as the work goes on.
What this is.
One 3D world of the coast that became Chennai, built across its eras from a verified evidence graph, so that a poster, a short film, an explorable website and eventually a walk through the old city are all renders from the same place. Geography, dates and named structures are source-locked. Textures, counts and unnamed forms are illustrative and say so in the record.
I am not a historian. I was born in Tamil Nadu, I live in San Francisco, and I build in AI. I have always liked history and I have wanted to work in 3D animation since I was a child and never got there. The models made both possible at once: a way to read the sources faster, and a way to build the thing the sources describe. So this is a learning project as much as a making one.
It is not one-shot prompting. It has taken hours of patience, being the human in the middle, going back and forth, patching, measuring, throwing things away. Two models do the work, one planning and judging and one building, and every phase ends with a written judgment against a visual target and a historical one. Nothing is approved by being rendered. I open each phase and I accept each report, and I get things wrong in public so they can be corrected.
Chennai is harder than a Western city to build this way. There are hundreds of reference points for a European street from any century. For this coast in the 1630s or the 1700s there are a handful of plans, some letters, and Love. Most of the photographed evidence begins in the 1900s. That is part of why it is worth doing.
Where it stands.
| Phases built and judged | 14 |
|---|---|
| Look against the target, 0 to 10 | 8.0 |
| Historical conformance, 0 to 10 | 9 |
| Placed instances in the world | 29,030 |
| Evidence graph | 340 nodes, 189 edges |
| Paid generation jobs, all recorded | 313 |
| Generation spend so far, rough | about $110 |
| Model compute so far at list prices, rough | about $2,700 |
| On disk | 34 GB of work plus 15 GB of version history, 27,442 files |
| Now | phase 15 briefed: the walkthrough, one recorded pass from the aerial views down to the beach and the lane across the three states, plus three bounded repairs (one man's skin, the far town drawn as cards, a frame budget measured on a quiet machine) |
Who.
Vinay Chandrasekaran. Tamil Nadu-born, based in San Francisco. This is a hobby, not a product and not my startup. If a line is wrong, say so. A source I missed is more useful than a general scolding. Corrections: message me on LinkedIn.