Making a consumer interface that visualizes a real-world construction — just like the Thirty Meter Telescope’s mirror — would possibly appear to be a job that calls for deep data of geometry, D3.js, and SVG graphics. However with a Massive Language Mannequin (LLM) like Claude or ChatGPT, you needn’t know every part upfront.
This text paperwork a journey in constructing a posh, interactive UI with no prior expertise in D3.js or UI growth usually.
The work was accomplished as a part of constructing a prototype for an operational consumer interface for the telescope’s main mirror, designed to indicate real-time standing of mirror segments.
It highlights how LLMs assist you “get on with it”, providing you with a working prototype
even whenever you’re unfamiliar with the underlying tech.
Extra importantly, it reveals how iterative prompting — refining your requests step-by-step —
leads not solely to the precise code but in addition to a clearer understanding
of what you are attempting to construct.
The Objective
We wished to create an HTML-based visualization of the Thirty Meter Telescope’s main mirror, composed of 492 hexagonal segments organized symmetrically in a round sample.
We started with a high-level immediate that described the construction, however quickly realized that to succeed in my purpose, I might must information the AI step-by-step.
Step 1: The Preliminary Immediate
“I wish to create an HTML view of the Thirty Meter Telescope’s honeycomb mirror.
Attempt to generate an HTML and CSS primarily based UI for this mirror, which consists of 492 hexagonal segments organized in a round sample.
Total construction is of a honeycomb. The construction ought to be symmetric.
For instance the variety of hexagons within the first row ought to be similar within the final row.
The variety of hexagons within the second row ought to be similar because the one within the second final row, and many others.”
Claude gave it a shot — however the outcome wasn’t what I had in thoughts. The format was blocky and never fairly symmetric. That is after I determined to take a step-by-step method.
Step 2: Drawing One Hexagon
“This isn’t what I need… Let’s do it step-by-step.”
“Let’s draw one hexagon with flat edge vertical. The hexagon ought to have all sides of similar size.”
“Let’s use d3.js and draw svg.”
“Let’s draw just one hexagon with d3.”
Claude generated clear D3 code to attract a single hexagon with the proper orientation and geometry. It labored — and gave me confidence within the constructing blocks.
Lesson: Begin small. Affirm the muse works earlier than scaling complexity.
Step 3: Including a Second Hexagon
“Good… Now let’s add yet one more hexagon subsequent to this one. It ought to share vertical edge with the primary hexagon.”
Claude adjusted the coordinates, inserting the second hexagon adjoining to the primary by aligning their vertical edges. The format logic was starting to emerge.
Step 4: Creating the Second Row
“Now let’s add yet one more row.
The hexagons within the second row share vertical edges with one another just like the primary row.
The highest slanting edges of the hexagons within the second row ought to be shared with the underside slanting edges of the hexagons within the first row.
The variety of hexagons within the second row ought to be such that the primary row seems centrally positioned with the second row.”
Preliminary makes an attempt did not correctly align the slanting edges.
“Oops… this doesn’t share the slanting edges with the earlier row.”
However ultimately, after clarifying spacing and offset logic, Claude acquired it proper.
Lesson: Geometry-based layouts usually require a number of iterations with cautious visible inspection.
Step 5: Increasing right into a Symmetric Construction
“Now we have to create larger construction with extra hexagons organized in additional rows such that:
The general construction seems round like honeycomb.
The variety of hexagons within the rows goes on growing after which goes on lowering to type a wonderfully symmetric construction.
The entire variety of hexagons must be 492 to match the TMT telescope.
We will have an empty hexagon (exhibiting empty house) precisely on the heart of the circle.”
Claude used a ring-based format method to simulate round symmetry. However at first:
“This isn’t round however seems extra like a hexagonal total view…”
Then I instructed:
“Strive with solely 6 hexagons within the first and final row.”
This alteration improved symmetry and helped obtain a visually round format. The variety of hexagons per row elevated after which decreased — precisely as desired.
Step 6: Tuning the Central Opening
“That is higher however we’d like a smaller opening on the heart.The black house on the heart is simply too large. It ought to be at most 1 or a couple of hexagons.”
By lowering the empty house and rebalancing the interior rings, we lastly acquired a well-packed, round construction with a small central hole — matching the TMT design.
Lesson: Use domain-specific constraints (like whole depend = 492) as guideposts for format parameters.
Step 7: Including Numbering and Tooltips
“We wish to have a quantity on every hexagonal section. They need to be numbered sequentially. The primary within the first row ought to be 1 and the final within the final row ought to be 492. Once we present the hexagonal section data on mouseover, we must always present the quantity as properly.”
Claude initially assigned numbers primarily based on ring index, not row order.
“You’re producing numbers primarily based on place within the ring… However the numbering ought to be row-based. So we must always someway map the rings to the row. For instance, Ring 13 section quantity 483 is in row 1 and ought to be numbered 1, and many others. Are you able to recommend a technique to map segments from rings to rows this manner?”
As soon as this mapping was carried out, every part fell into place:
- A round format of 492 numbered segments
- A small central hole
- Tooltips exhibiting section metadata
- Visible symmetry from outer to interior rings
Reflections
This expertise taught me a number of key classes:
- LLMs assist you get on with it: Even with zero data of D3.js or SVG geometry, I may begin constructing instantly. The AI scaffolded the coding, and I realized via the method.
- Prompting is iterative: My first immediate wasn’t unsuitable — it simply wasn’t particular sufficient. By reviewing the output at every step, clarified what I actually wished and refined my asks accordingly.
- LLMs unlock studying via constructing: In the long run, I did not simply get a working UI. I acquired an comprehensible codebase and a hands-on entry level into a brand new know-how. Constructing first and studying from it.
Conclusion
What began as a imprecise design thought was a functioning, symmetric, interactive visualization of the Thirty Meter Telescope’s mirror — constructed collaboratively with an LLM.
This expertise reaffirmed that prompt-driven growth is not nearly producing code — it is about pondering via design, clarifying intent, and constructing your approach into understanding.
For those who’ve ever wished to discover a brand new know-how, construct a UI, or sort out a domain-specific visualization — do not wait to study all of it first.
Begin constructing with an LLM. You will study alongside the best way.
