Decoding Lighting Simulations with AI: A Real-World Experiment

Can ChatGPT make sense of a DIALux simulation? Join us as we test how AI simplifies the complex world of lighting reports and elevates design thinking.

Apr 28, 2025

Last week, I gave you an overview of how AI could potentially be used in lighting design — from interpreting design briefs, to visual ideation, to decoding simulation outputs, and more.

Here’s the earlier post in case you missed it:

Illuminated Intelligence: Where Lighting Design Meets Learning and the Algorithm

Sahil Tanveer

Apr 21

Illuminated Intelligence: Where Lighting Design Meets Learning and the Algorithm

The Shift in the Source

Read full story

This week, I want to zoom in on one aspect: where outputs from simulation tools can be simplified into strategic decisions to improve the efficiency of design.

If you’re a lighting designer who regularly uses simulation tools and photometric analysis, you’ll want to read this one closely.

With the help of real-world lighting design project simulation data — contributed generously by our friend and a veteran lighting designer, Martin Klaasen of Nulty+ and Klaasen Lighting Design — we ran an experiment:
We analyzed complex diagrams, luminaire choices, and photometric data using Language Models (primarily ChatGPT) for a particular interior space.

The simulation tool considered here is DIALux, one of the oldest and most broadly used platforms for lighting design and analysis.

Here’s what we did:

Upon receiving a DIALux Report for a particular project from Martin, we began by manually reading it to get a fair sense of what the report contained.
Of course, being architects, we could understand parts of it completely, some only partially, and some not at all!

This, in many ways, is the perfect place to start:
Inquiring, deciphering, and unpacking the parts we don't immediately understand — using the powerhouse that is ChatGPT.

Step 1: Plain Language Summary with ChatGPT 4o

First, we asked ChatGPT 4o to unravel the report and explain it in simple, non-technical language.
We used a fairly straightforward prompt, requesting a summary suitable for someone who is not a lighting designer.

Here’s what it gave me:

🔹 Overall Report Summary

The document analyzes the lighting design and performance for a facility comprising various types of spaces like washrooms, offices, circulation areas, and storage zones.

Lighting type used throughout:
➔ 36W ECO LED battens, 4ft length, 4000K (cool white light).
Performance per light:
➔ Each luminaire produces about 4300 lumens, delivering high energy efficiency.

🔹 Space-wise Lighting Overview

Each room or area has been assessed for average lighting levels to check whether it meets typical functional requirements.

🔹 Key Observations

Uniformity Problems:
Some spaces, especially the Male Ablution and small office zones, show high contrast between bright and dark patches. This can lead to a patchy lighting experience.
Underlit Office Zones:
The open office and meeting room areas have lighting levels around 190 lux. Ideally, modern offices and meetings spaces should aim closer to 300–500 lux for visual comfort and task performance.
Well-performing Areas:
The Crib (canteen) area and Female Ablution have good brightness and decent uniformity.
Luminaire Note:
The luminaires are efficient, but detailed photometric reliability is slightly limited because the manufacturer is not officially certified in the DIALux database.

🔹 Technical Terms Simplified

Lux (lx) = How bright a surface appears under lighting.
Uniformity (Uo) = How evenly the light is spread (closer to 1 = better).
Emin/Emax = The range between darkest and brightest points measured.

🔹 Architectural Advice

For offices and meeting rooms ➔ Recommend adding a few more light fittings or upgrading the output per fitting to improve working conditions.
For areas with low uniformity ➔ Suggest minor layout adjustments or additional fittings to avoid visible patches of uneven light.

What you see above is as good as an overview or general understanding of the report achieved with basic prompting.
It might feel like enough to act upon, but trust me — it isn’t.

This is just a launchpad, not an actionable basis for any design decisions.

Step 2: Deeper Reasoning with ChatGPT o3

Next, we opened a new chat with ChatGPT o3 to inject some reasoning into the inferences drawn from the simulation report.

Since ChatGPT now has memory across chats, o3 referenced the earlier conversation we had with 4o about the DIALux report.
We asked for an overview again, this time requesting slightly deeper insight.

Here’s what o3 said:

What the Original DIALux File Tells Us — at a Glance

Document metadata
- Prepared in May 2024 with current DIALux release.
- One luminaire family used everywhere: a 36 W linear LED batten delivering ≈ 4,300 lm (≈ 119 lm/W).
- Totals – ≈ 70 fittings, 2520 W connected load, ≈ 301 000 lm emitted.

1 Luminaire Schedule & Photometric Sheet

Basic electrical data, CCT (~4000 K), CRI = ≥ 80, polar curve.
No dimming interface shown (fixed-output gear).

2 Room-by-Room Calculation Tables

(All glare proxy values g₁ / g₂ lie between 0.07 – 0.41, i.e. within comfortable range.)

3 Layout Plans

Each space includes a ceiling plan with fitting coordinates and mounting height (2.4 m in washrooms, 2.7 m elsewhere).
Larger rooms rely on simple rectangular grids; smaller rooms use isolated single rows.

4 Energy Snapshot

Overall Lighting Power Density ≈ 10–11 W/m² (derived from area implied by grids).
Acceptable for industrial-office benchmarks but achieved without any control strategy.

Now that we had set a stronger foundation, we moved on to the real fun — a deeper critical analysis of the report.

We asked for targeted and specific improvement suggestions for the lighting design.

IMP NOTE:

These suggestions are aimed at professional lighting designers with a highly trained, critical eye. They do not replace the valuable expertise of human professionals.

🚨 Don’t try this at home! 🚨

Step 3: Prompt Engineering for Decision-Ready Insights

At this point, engineering the right prompt became critical.
If you’re new to prompt engineering for architecture and design, you might want to check out our 6-part crash course here:

Prompt Engineering Crash Course

We structured our prompt around the four essential aspects of good prompting:

Persona (Who are you asking it to be?)
Task (What do you want it to do?)
Context (What is the situation or background?)
Format (How do you want the answer?)

With this structured approach, we decoded the report in detail, leading to a highly informed and clear understanding, ready for design decisions.

Here’s the final inference we got from ChatGPT o3:

Lighting Performance Review & Improvement Road-Map

Master Oogway, here is my deep-dive into the current calculation file. I read the report several times, extracted every illuminance table and luminaire layout, and benchmarked the data against the values prescribed in EN 12464-1 (comparable to IS 3646 Part II) and the IESNA Lighting Handbook. Below I first explain — transparently and step-by-step — how I arrived at each inference, then I list the strengths, latent opportunities and precise corrective actions for every space, and finally I wrap everything up in an executive summary you can forward to the project team.

[Side note: My ChatGPT calls me Master Oogway 🐢 and I call it Skye 🩵]

1 Methodology — how I interrogated the DIALux output

2 Room-by-Room Assessment & Action List

Shift the two perimeter fittings 300 mm inward to equalise the overlap.
Commission sensor-dimming as above.| |Open Office|Ē = 192 lx; Uo = 0.42 |Target for screen-based tasks is 300–500 lx / ≥ 0.6 → fails both.|
- Increase row count from 2 to 3 (3 × 3 grid).
Alternatively, replace current battens with higher-output 6,000 lm versions.
Paint ceiling reflectance > 0.7 to recover ~10 % upward flux.| |Meeting Room|Ē = 193 lx; Uo = 0.42 |Needs 300 lx & 0.6 → fails.|
• Introduce a pendant direct/indirect fitting above the table (3,000 K to aid alertness).
• Keep existing battens on a second circuit for “presentation” mode.| |Pre-Start Open Area (Circulation)|Ē = 277 lx; Uo = 0.47 |For corridors only 100 lx required – overlit by ~175 %.|• Remove every fourth fitting (maintains ≈180 lx).
• Link remaining grid to occupancy control; pay-back < 1 year.| |Crib|Ē = 377 lx; Uo = 0.47 |Target 300 lx / 0.6 → level OK, uniformity moderate.|• Rotate every alternate batten 90° to spread lateral light.
• Consider warm-white (3,000 K) linear under-cabinet strips for visual comfort.| |Ice Room (Cold Store)|Ē = 248 lx; Uo = 0.56 |Target 200 lx / 0.4 – Pass.|• Fit IP66 anti-corrosion housings (DIALux file shows standard troffer).| |Permit Office|Ē = 284 lx; Emin = 67 lx; Uo = 0.24 |Falls short on uniformity (needs ≥ 0.6).|1. Reduce spacing to 2.4 m or add a central row (6-pack).
Install desk-mounted task lights to raise local illuminance without boosting ceiling load.| |Energy & Controls (whole site)|Total 2 520 W, ≈11 W/m² (estimated).|Good vs. ASHRAE 90.1 (13 W/m² cap for industrial office).|• Site-wide occupancy sensing could cut burn-time by 30–40 %.
• Daylight dimming in open-office facade zones would save a further 10 %.|

3 Advantages Already Embedded in the Design

LED efficacy (119 lm/W) keeps the design future-proof for most energy codes.
Consistent 4 000 K CCT throughout gives uniform visual appearance and reduces inventory complexity.
Low glare proxies (g1 ≤ 0.34) suggest user comfort has been considered.
Straight batten geometry simplifies maintenance and future retro-fits.

4 Improvement Matrix (priority → effort)

5 Consequences of not acting

Under-lit office areas can cause eye strain, reduce productivity by up to 8 %.
Patchy illumination (Uo < 0.3) triggers CCTV exposure swings and visible flicker in security footage.
Over-lit corridors waste ≈ 1 MWh/year – equivalent to ~800 kg CO₂.
No controls → lamps run 24/7; L₇₀ life will halve, pulling relamping forward by 2–3 years.

6 Executive Summary

Bring every task area up to 300 lx & ≥ 0.6 uniformity.
Remove excess light in low-priority zones and redeploy the wattage budget to task zones.
Layer controls – presence/absence detection in sanitaries & circulation, daylight harvest at facades, dual-scene switching in presentation spaces.
Standardise on a next-generation ≥ 150 lm/W batten for all future orders to lock in energy savings without redesign.
Commission & measure – use spot-lux readings post-install to verify targets, then fine-tune dim-levels in the control system.

Implementing the high-impact items (1 + 2 + 3) will:

Raise user comfort scores immediately,
Trim annual lighting energy by an estimated 35 %, and
Extend luminaire lifetime thanks to lower drive currents and fewer burning hours.

After the entire exercise, we had Martin review it, of course! Here’s what he had to say:

To receive such an amazingly detailed report by itself is quite astonishing. On the surface, it captures the process and the most relevant and appropriate details correctly. Considering that what you see is what we got, amazing! There are a few words, definitions, terminology usages and references that can be fine-tuned or would probably be slightly different. Some of the calculations referred to seem correct at first glance, but that will take even a professional like me quite a bit of time to comb through and check. I don't have the computing and reading speed of AI! As a first shot to see how far we can push AI to assist us in the design process, this is remarkable! Can’t wait to push this to the next level. Oh, and yes, this is still very "factual"; the human aspect is still missing or hardly touched upon. But then we started with calculations for a project site office, a mostly functional space... :)

That’s it for this week, folks!
If you want to learn the foundations of AI for Lighting Design, I’m thrilled to announce that we’re working on releasing a foundational AI course for Lighting Designers, in collaboration with one of the leading global experts, Martin Klaasen of Klaasen Lighting Design and Nulty+.

While we quietly work behind the scenes to bring you the best learning experience — combining Martin’s 40+ years of lighting expertise with my work in AI in Architecture and Design — here’s a sneak peek of what’s coming.

If this feels like something you’ve been searching for:

Join The Waitlist

AI: Fundamentals For Lighting Designers

36.3MB ∙ PDF file

Download

Download the Brochure to know more about this Introductory Online Course, led by Ar Sahil Tanveer and Martin Klaasen, introducing AI’s transformative role in concept development, visualization, and photometric analysis. Through structured lessons and hands-on exercises, participants will explore AI-driven workflows and intelligent lighting strategies—whether they’re new to AI or refining their expertise.