The document discusses the role of large language models (LLMs) in enabling end-users to customize software code, particularly through the transformation of high-level specifications into variable code for various applications. It highlights two scenarios where LLMs assist non-professional programmers in implementing variability and constructing configurable tools. The paper emphasizes the potential for LLMs to enhance software customization while acknowledging challenges and advocating for future research in this area.

1. A Demonstration of End-User
Code Customization Using
Generative AI
Mathieu Acher
@acherm

2. Preprint
https://hal.science/hal-04312909
https://vamos2024.inf.unibe.ch/cfp/
Presented at 18th International Working Conference on Variability Modelling of
Software-Intensive Systems (VaMoS 2024) in Bern, Switzerlan
2

3. Generative AI and variability: rising interests!
Mathieu Acher and Jabier Martinez. 2023. Generative AI for Reengineering Variants
into Software Product Lines: An Experience Report. VariVolution@SPLC 2013
Mathieu Acher, Jean-Marc Jézéquel, and José A. Galindo. 2023. On Programming
Variability with Large Language Model-based Assistant. In SPLC 2023
Sandra Greiner, Klaus Schmid, Thorsten Berger, Sebastian Krieter, Kristof Meixner
Generative AI And Software Variability – A Research Vision, VaMoS 2024
Galindo et al. Large Language Models to generate meaningful feature model
instances, SPLC 2023
3

4. LLM
Hypothesis: Large language models (LLMs) act as a new variability compiler capable
of transforming a high-level specification (“prompt”) into variable code, features,
generators, configurable systems, etc. written in a given technological space.
Motto:
“features as
prompts”
4
Mathieu Acher, Jean-Marc
Jézéquel, and José A.
Galindo. 2023. On
Programming Variability
with Large Language
Model-based Assistant. In
SPLC 2023

5. LLM
“End-user software engineering focuses on empowering individuals
who are not professional programmers to design, implement, and
maintain software applications.“
Two scenarios:
● (1) end-users can implement variability with the assistance of LLMs;
● (2) LLMs can be used to construct specialized, configurable tools for
end-users
Motto:
“features as
prompts”
5

6. 6

7. 7

8. 8
What’s this TikZ code is doing?
Code understanding

9. 9
What’s this TikZ code is doing?
Code understanding

10. 10
What’s this TikZ code is doing?
Code understanding
The level of details is too low, and
the explanations are not really
actionable, especially for relating
code elements to visual aspects.

11. 11

12. 12
“can you vary the eye of the squid
with the blue color? tell
me what to varies in the TikZ code”
Feature location
To change the color of the squid’s eye to blue, you’ll need
to modify the fill commands that define the eye. The lines
that define the eye are:
fill [Grey50]
(64, 16) circle [radius=32];
fill [BlueGrey900] (64, 16) circle [radius=12];
To change the eye color to blue, you can replace the
colors
...
fill [blue] (64, 16) circle [radius=32];
fill [blue!60!white] (64, 16) circle [radius=12];
...

13. 13
“would it be possible to vary the
original code using a
templating engine in Python in such
a way it is possible to have
either eye of the squid with the blue
color or the original eye
(as in the original TikZ code)? ... can
you parameterize the spots
with a random color? if not random
mode, then it’s the default
color”
Feature and variation
implementation

14. 14
Combination of features
and automated
derivation.

15. 15
Beyond TikZ (feature
location,
implementation, etc.)

16. Here is a TikZ code:
tikzset{%
cat/.pic={
tikzset{x=3cm/5,y=3cm/5,shift={(0,-1/3)}}
useasboundingbox (-1,-1) (1,2);
fill [BlueGrey900] (0,-2)
.. controls ++(180:3) and ++(0:5/4) .. (-2,0)
arc (270:90:1/5)
.. controls ++(0:2) and ++(180:11/4) .. (0,-2+2/5);
foreach i in {-1,1}
scoped[shift={(1/2*i,9/4)}, rotate=45*i]{
clip [overlay] (0, 5/9) ellipse [radius=8/9];
clip [overlay] (0,-5/9) ellipse [radius=8/9];
fill [BlueGrey900] ellipse [radius=1];
clip [overlay] (0, 7/9) ellipse [radius=10/11];
clip [overlay] (0,-7/9) ellipse [radius=10/11];
fill [Purple100] ellipse [radius=1];
};
fill [BlueGrey900] ellipse [x radius=3/4, y radius=2];
fill [BlueGrey100] ellipse [x radius=1/3, y radius=1];
fill [BlueGrey900]
(0,15/8) ellipse [x radius=1, y radius=5/6]
(0, 8/6) ellipse [x radius=1/2, y radius=1/2]
{[shift={(-1/2,-2)}, rotate= 10] ellipse [x radius=1/3, y radius=5/4]}
{[shift={( 1/2,-2)}, rotate=-10] ellipse [x radius=1/3, y radius=5/4]};
fill [BlueGrey500]
(-1/9,11/8) ellipse [x radius=1/5, y radius=1/5]
( 1/9,11/8) ellipse [x radius=1/5, y radius=1/5];
fill [Purple100]
(0,12/8) ellipse [x radius=1/10, y radius=1/5]
(0,12/8+1/9) ellipse [x radius=1/5 , y radius=1/10];
foreach i in {-1,1}
scoped[shift={(1/2*i,2)}, rotate=35*i]{
clip [overlay] (0, 1/7) ellipse [radius=2/7];
clip [overlay] (0,-1/7) ellipse [radius=2/7];
fill [Yellow50] ellipse [radius=1];
};
scoped{
clip (-1,-2) rectangle ++(2,1);
fill [BlueGrey900] (0,-2) ellipse [radius=1/2];
fill [Grey100]
(-1/2,-2) ellipse [x radius=1/3, y radius=1/4]
( 1/2,-2) ellipse [x radius=1/3, y radius=1/4];
};
foreach i in {-1,1}
foreach j in {-1,0,1}
fill [Grey100, shift={(0,11/8)}, xscale=i, rotate=j*15,
shift=(0:1/2)]
ellipse [x radius=1/3, y radius=1/64];
},
I’d like to change the color of the cat’s tail in red
inaccurate
feature location
or/and incorrect
code synthesis
is totally
possible (even
given the same
prompt)
Feature location:
beware!

17. To change the color of the squid’s eye to blue, you’ll need
to modify the fill commands that define the eye. The lines
that define the eye are:
fill [Grey50]
(64, 16) circle [radius=32];
fill [BlueGrey900] (64, 16) circle [radius=12];
To change the eye color to blue, you can replace the
colors
...
fill [blue] (64, 16) circle [radius=32];
fill [blue!60!white] (64, 16) circle [radius=12];”
...
“can you vary the eye of the squid
with the blue color? tell
me what to varies in the TikZ code”
Lots of back and
forth for
integrating the
code!

18. “can you vary the eye of the squid
with the blue color? tell
me what to varies in the TikZ code”
“would it be possible to vary the
original code using a
templating engine in Python in such
a way it is possible to have
either eye of the squid with the blue
color or the original eye
(as in the original TikZ code)? ... can
you parameterize the spots
with a random color? if not random
mode, then it’s the default
color”
Prompts flow: tension
between high-level
specification and
implementation details

19. Conclusion
LLMs can be leveraged to support end-users in the customization of code and
technical artefacts (customizing TikZ without knowing TikZ?)
Transformation of an undocumented, technical, low-level TikZ into a user-friendly,
configurable, Web-based customization tool written in Python, HTML, CSS, and
JavaScript and itself customizable.
LLMs can support traditional variability engineering tasks, such as identification and
implementation of features, synthesis of a template code generator, and development
of end-user configurators.
The rise and fall of LLMs: many caveats and failures at the different steps!
Future work: thorough evaluation and more robust, automated, end-to-end
customization
Exciting time: democratization of programming and software-related customization! 19

20. LLM4Code
many open positions at DiverSE
please contact me!
20