Archit Sunat All work

A NASA spreadsheet, turned into somewhere you can fly

Exoplanet Universe

An interactive 3D map of 6,150 confirmed exoplanets, built from the CSV NASA publishes and covering thirty years of discovery from 51 Pegasi b in 1995 to the TESS era. Live, public, and free to fly around.

Visit it →

The real thing, running here. Press Launch, then drag to look and scroll to fly. Every planet is a row from the NASA archive. Open it full screen →
01

Where it started

NASA has the best data. The interaction design was the problem.

I spend time in NASA's Eyes on Exoplanets, the reference tool, built on the most accurate catalogue that exists. Every planet is a star-shaped sprite; zoom in and you get an ellipse and a label. It is completely accurate and I have never felt much looking at it. None of that is a data problem. It is a visualisation and interaction design problem, which is what this project is.

The archive gives a row per planet, and right ascension, declination and distance are already a point in space. Distance is the interesting part: the nearest planet is four light-years out and the furthest nearly thirty thousand, so plot it honestly and everything except the far edge collapses into a dot. Distance is compressed logarithmically, and the app tells you where you are in light-years the whole time you are moving, so the compression is never hidden.

One archive row converted into a position, with distance compressed logarithmically ONE ROW OF THE CSV ra 285.679dec +50.241dist_ly 2,545radius 1.63period 384.8star_temp 5757 A POINT IN SPACE Right ascension and declination point at it. Distance says how far along that line it sits. Direction is exact. Distance is ordered, not literal. 4 40 400 4,000 30,000 LIGHT-YEARS, COMPRESSED SO THE NEAR NEIGHBOURHOOD STAYS EXPLORABLE
02

The product calls

What to render, and what the surfaces should look like.

The obvious build gives every planet a sphere, a texture and a label, and a page that takes half a minute to load and runs at nine frames a second. So they are not all the same kind of object: a hundred and thirty-seven get real meshes and information panels, the other six thousand are one point cloud drawn in a single pass, and clicking any of them promotes it on the spot.

No telescope can resolve an exoplanet, so there are no textures to find and stock planet art would be a lie. The surface is generated from what NASA does know: temperature picks the world, above 1,600 kelvin lava, above 350 a Venus-like haze, below 140 frozen nitrogen, and the planet's own identifier seeds the tint.

CFHTWIR-Oph 98 b, 646 light-years out. The surface is generated, and the app says so on the panel next to it.
CFHTWIR-Oph 98 b, 646 light-years out. The surface is generated, and the app says so on the panel next to it.
TRAPPIST-1, seven planets around one star, with the ten light-year ring behind it for scale.
TRAPPIST-1, seven planets around one star, with the ten light-year ring behind it for scale.

The camera was the hardest part. A fixed zoom step is unusable across this range, so every notch moves you a percentage of the distance you have left and you approach a surface without quite reaching it. Click any planet and it flies there itself and arrives framed.

03

Evals

How do you check that a hundred and thirty-seven planets look right?

A 3D visualisation can pass every unit test and still put a planet in the wrong place, in the wrong colour, or with a confidently wrong number under it.

The build ran as a Driver and a Navigator: one agent writes and is not allowed to decide what to build, the other reviews and never writes. An agent reviewing its own output agrees with itself every time. Writing the spec is the part that stayed mine, because the Navigator can only review against something, and that is a limit on me before it is one on them.

About

Most of the seven years was spent alongside the CEO, the head of design, engineering, business analysts, subject matter experts and growth.

What I want next is AI-native product work, and a place where I can keep building great products myself.

Education

  • MS, Cognitive Science

    Indian Institute of Technology Gandhinagar2015 to 2017

    Thesis on the cognitive system of musical improvisation, read through phenomenology.

  • BS, Life Science and Biochemistry

    St. Xavier's College, Mumbai2012 to 2015