Resea ch Pape
Recommended ci a ion: Ga cia Hue es, S., & B agós, R. (2025). A Hyb id
Assessmen App oach o AI-Enhanced Challenge-Based Lea ning in Enginee ing
Educa ion. In Kangaslampi, R., Langie, G., Jä inen, H.-M., & Nagy, B. (Eds.),
SEFI 53 d Annual Con e ence. Eu opean Socie y o Enginee ing Educa ion
(SEFI), Tampe e, Finland. DOI: 10.5281/zenodo.17631282.
This Con e ence Pape is b ough o you o open access by he 53 d Annual Con e ence
o he Eu opean Socie y o Enginee ing Educa ion (SEFI) a Tampe e Uni e si y in
Tampe e, Finland. This wo k is licensed unde a C ea i e Commons
A ibu ion-NonComme cial-Sha e Alike 4.0 In e na ional License.
A HYBRID ASSESSMENT APPROACH TO AI-ENHANCED
CHALLENGE-BASED LEARNING IN ENGINEERING EDUCATION
S. Ga cia-Hue es a,1, R. B agós b
a Telecos-BCN, Uni e si a Poli ècnica de Ca alunya (UPC), Ba celona, Spain,
ORCID 0000-0002-6735-3025
b Telecos-BCN, Uni e si a Poli ècnica de Ca alunya (UPC), Ba celona, Spain,
ORCID 0000-0002-1373-1588
Con e ence Key A eas: Digi al ools and AI in enginee ing educa ion; Building he
capaci y and s eng hening he educa ional compe ences o enginee ing educa o s
Keywo ds: Gene a i e AI (GenAI); Challenge-Based Lea ning; Hyb id Assessmen
ABSTRACT
This pape in oduces a hyb id assessmen app oach in eg a ing Gene a i e AI
(GenAI), speci ically Cha GPT, in o challenge-based enginee ing educa ion.
S uden s i e a i ely de ined p oblem s a emen s, b ains o med solu ions, and
p esen ed hei wo k, ecei ing eedback h ough ub ic-based GenAI assessmen s
o Cha GPT in e ac ions and di ec acul y e alua ion. While Cha GPT posi i ely
in luenced s uden pe o mance and in ellec ual engagemen , i ampli ied exis ing
di e ences in mo i a ion, p io knowledge, skills, and eam dynamics, unde sco ing
he impo ance o comp ehensi e, human-cen e ed pedagogical app oaches.
Compa ed o p e ious i e a ions, his app oach acili a ed obus hypo hesis
alida ion and meaning ul s uden e lec ion on AI usage. Con a y o conce ns abou
educed e o , mos eams le e aged Cha GPT o deepe explo a ion and c i ical
hinking, esul ing in mo e con iden p esen a ions and highe -quali y con en . These
indings ein o ce e idence ha explici GenAI adop ion, wi h guided acul y
o e sigh , suppo s design- hinking s a egies, enhances s uden engagemen , and
main ains academic in eg i y. Con inued me hodological e inemen and u he
empi ical esea ch emain essen ial o maximize GenAI’s educa ional bene i s in
challenge-based enginee ing educa ion.
1 Co esponding Au ho
S. Ga cia-Hue es
[email p o ec ed]
1 INTRODUCTION
The ise o Gene a i e AI (GenAI) ools, pa icula ly la ge language models such as
Cha GPT, has a ac ed conside able schola ly a en ion and deba e wi hin highe
educa ion, no ably in enginee ing educa ion. His o ically, enginee ing educa ion
con inuously adap s o echnological ad ancemen s, p omp ing shi s in eaching
me hods and lea ning s a egies. A clea his o ical pa allel exis s wi h he
in oduc ion o calcula o s in o ma hema ics class ooms du ing he la e 20 h cen u y.
Ini ially, calcula o s aced skep icism om educa o s ega ding hei po en ial
nega i e impac on s uden s' analy ical skills. Howe e , esea ch by Thomas e al.,
(2006) shows calcula o s e en ually gained accep ance, illus a ing how ini ial
esis ance o en p ecedes in eg a ion o ans o ma i e educa ional echnologies.
Today, enginee ing educa o s a e simila ly challenged by he apid eme gence o
GenAI ools, necessi a ing p oac i e and open in eg a ion in o educa ional p ac ice.
Recen s udies unde sco e he g owing consensus ha , despi e ini ial e hical and
academic in eg i y conce ns, he p og essi e in oduc ion o GenAI AI in o highe
educa ion cu icula is bo h ine i able and bene icial, signi ican ly enhancing
pedagogical s a egies, cu iculum de elopmen , and s uden engagemen (San os
e al., 2024; S anko ski e al., 2024). Fu he mo e, con empo a y case s udies e eal
how embedding Gen AI wi hin s uc u ed educa ional amewo ks, such as design
hinking cu icula, subs an ially accele a es inno a ion p ocesses and imp o es
ou comes o bo h s uden s and educa o s (Squalli Houssaini e al., 2024).
The e o e, he p esen s udy p oposes a clea and s uc u ed me hodology explici ly
encou aging acul y and s uden s o clea ly adop GenAI ools wi hin
challenge-based enginee ing cou ses. By p o iding sha ed access o Cha GPT
accoun s and le e aging he analy ical capabili ies o hese ools hemsel es o
quali a i e and quan i a i e assessmen , we aim o enhance s uden s' c ea i i y and
i e a i e design skills, while simul aneously p o iding acul y wi h a deepe
unde s anding o s uden in e ac ions. The o e a ching esea ch objec i e add essed
in his s udy is: "How can he anspa en in eg a ion and sha ed acul y-s uden use
o GenAI ools like Cha GPT e ec i ely enhance bo h lea ning p ocesses and
assessmen quali y in challenge-based enginee ing educa ion?"
2 METHODOLOGY
The ac i i y aims o enhance he En ep eneu ship and Inno a ion o Wo ld
Challenges (EIWC) cou se wi hin he Elec onic Enginee ing Mas e a Uni e si a
Poli ècnica de Ca alunya (UPC). S uc u ed as a Challenge-Based (Kohn Rådbe g e
al., 2020) Lea ning cou se (5 ECTS, 3 hou s / week), i emphasizes socie al impac ,
inco po a ing speci ic sessions on sus ainabili y and e hics. Teams engage in h ee
p og essi ely complex challenges. The ini ial wo-week challenge allows s uden s o
le e age amilia me hods and ecognize me hodological gaps. In he second
six-week challenge, he eaching eam in oduces Design-Thinking (Lei e & Meinel,
2016) and Lean-S a up (Ries, 2011) amewo ks a e in oduced by he eaching
eam, each sho p esen a ion ollowed by a hands-on ac i i y pe o med by he
eams. In he hi d challenge, a dis up i e echnology, in a low phase o he TRL
scale is p esen ed by a s a up o esea ch g oup and he eams should p opose
al e na i e applica ions and business models, in ended o ha e socie al impac .
The ac i i y desc ibed in his pape ook place in he second challenge, speci ically in
he Need inding and Idea ion phases. Wi h he p e ious app oach, a 3 hou session
(wi h he co esponding au onomous p epa a ion wo k in he p eceden week) was
de o ed o Need inding, esul ing in a challenge b ie iden i ying a speci ic and
alida ed need, na owe han he o iginal challenge and he second session o
idea ion, esul ing in a solu ion concep , selec ed om a se o al e na i es. The
ollowing sessions a e de o ed o he e inemen o he idea and o he de elopmen
o he Value P oposi ion and he Business Model, using he Business Model Can as
(Os e walde & Pigneu , 2010). In he las wo e ms, a non-homogeneous use o
GenAI by eams has been obse ed, p ima ily in ended o educe wo kload a he
han o imp o e esul s.
To clea ly a icula e he in eg a ion o GenAI wi hin he second challenge, ou
me hodology was s uc u ed ac oss h ee phases: P e-Challenge P epa a ion,
In-Session Facili a ion, and Pos -Challenge E alua ion as depic ed in Figu e 1.
Figu e 1. Timeline o he di e en phases o p epa e, execu e and e alua e all he ac i i ies
in ol ed in he desc ibed me hodology.
2.1 P e-Challenge P epa a ion
In p epa a ion, acul y de ined an in en ionally b oad and open-ended challenge on
pedes ian sa e y isks associa ed wi h dis ac ed walking in u ban en i onmen s.
This in en ionally b oad aming aimed o acili a e explo a o y hinking and i e a i e
e inemen o p oblem de ini ions.
Co e digi al in as uc u e was es ablished by c ea ing sha ed Cha GPT accoun s
linked o gene ic emails accessible by bo h s uden s and acul y. This acili a ed
anspa en GenAI in e ac ions and con inuous o e sigh . The explici aim was o
c ea e an en i onmen whe e s uden s could openly engage wi h Cha GPT, while
acul y e ained isibili y and could o e imely o ma i e eedback h oughou he
cou se du a ion o e en ually analyse he logs in he e alua ion pe iod.
The p o ided Cha GPT accoun was s ongly ecommended o acili a e consis en
moni o ing and compa a i e assessmen ac oss eams. Howe e , i s use was no
en o ced as manda o y, aiming o p ese e s uden au onomy and lexibili y in ool
selec ion. This decision had implica ions o da a consis ency, occasionally
complica ing compa a i e analysis due o a ied ex e nal ool use, as u he
add essed in he limi a ions sec ion.
2.2 In-Session Facili a ion
Cou se ac i i ies comp ised wo main class oom sessions complemen ed by
independen s uden wo k. The i s session began wi h an in oduc ion o
ounda ional design hinking concep s, clea ly di e en ia ing he need inding
(p oblem explo a ion) and idea ion (solu ion de elopmen ) phases. Teams hen used
Cha GPT i e a i ely o e ine and de ine hei in e p e a ion o he b oad challenge,
s uc u ing in e ac ions as comple e exchanges (s uden p omp s wi h Cha GPT
esponses), g ouped in o sequences called i e a ions o p og essi e e inemen .
The second session expanded hese echniques, p o iding dedica ed ime o
solu ion e inemen . Facul y con inuously moni o ed Cha GPT in e ac ions, o e ing
ailo ed o ma i e eedback. Final p esen a ions, deli e ed o ally wi h slide suppo ,
we e independen ly e alua ed by h ee acul y membe s, who hen consolida ed hei
assessmen s in o inal g ades. Ad anced ou pu s (e.g., p o o ypes, ma ke analyses,
business models) we e ecognized as e idence o deepe GenAI in eg a ion aligned
wi h cou se objec i es.
2.3 Pos -Challenge E alua ion
The e alua ion phase combined quali a i e and quan i a i e analyses o GenAI
usage. Quali a i e e alua ion was pe o med by le e aging one o he OpenAI
models, speci ically model o1, applying a de ailed ub ic (see Appendix) o assess
i e a i e p omp ing, c ea i i y, c i ical hinking, and e lec i e engagemen om
in e ac ion logs. Quan i a i ely, a Py hon sc ip sys ema ically analyzed hese logs,
measu ing me ics such as o al okens exchanged, con e sa ion coun s, and
a e age oken densi y pe i e a ion, ollowing es ablished analy ical p ac ices om
ecen esea ch (Al es & Cip iano, 2024).
This hyb id assessmen app oach pu pose ully inco po a ed GenAI in o bo h lea ning
and e alua ion p ocesses, o e ing acul y aluable addi ional insigh s in o s uden
in e ac ions. Fu he mo e, i empowe ed s uden s o c ea i ely le e age GenAI, while
ein o cing c i ical hinking and e lec ing on i s usage.
3 RESULTS
This sec ion p o ides an in-dep h analysis, p esen ing bo h quan i a i e me ics and
quali a i e e alua ions o s uden in e ac ions wi h Cha GPT, alongside de ailed
assessmen s o he inal p esen a ions deli e ed by each eam. The analysis is
s uc u ed a ound clea checkpoin s o illus a e changes and de elopmen pa e ns
in s uden engagemen , c i ical hinking, and he complexi y o hei i e a i e
in e ac ions. Addi ionally, insigh s de i ed om compa a i e e alua ion o eam
p esen a ions allow o a deepe unde s anding o he speci ic con ibu ions o GenAI
ools o s uden pe o mance.
3.1 Cha GPT usage
Each eam's usage o Cha GPT was analyzed a wo checkpoin s: an in e media e
poin (p io o he second wo king session) and a inal poin (immedia ely be o e he
p esen a ion). Table 1 summa izes bo h quan i a i e and quan i a i e me ics.
Table 1. Cha GPT Quan i a i e and Quali a i e Usage Assessmen (G een indica es an
inc ease; o ange indica es a dec ease a he inal checkpoin .)
In e media e
Final
Team
I e a ions
Token / i e
Usage e al
I e a ions
Token / i e
Usage e al
A
20
655
2,75
47
708
3,00
B
14
442
2,25
40
369
2,75
C
14
584
2,50
10
499
3,00
D
12
640
2,75
14
647
3,00
E
10
378
3,25
24
312
3,50
F
10
461
3,25
22
600
3,50
G
2
9
1,00
14
672
3,00
The quan i a i e assessmen indica ed clea imp o emen be ween he in e media e
and inal e alua ion phases ac oss mos eams. Wi h he excep ion o Team C, all
eams no ably inc eased hei in e ac ions wi h Cha GPT, highligh ing g ea e
amilia i y and con idence in u ilizing he ool o i e a i e p oblem explo a ion. Teams
A and B pa icula ly s ood ou by subs an ially inc easing hei numbe o i e a ions,
e lec ing in ensi ied engagemen wi h he GenAI ool. Meanwhile, he
okens-pe -i e a ion me ic e ealed ha mos eams main ained consis en
in e ac ion complexi y, implying sus ained, hough ul engagemen .
Quali a i ely, eams ini ially u ilized Cha GPT p ima ily o p elimina y b ains o ming
and p oblem cla i ica ion. Howe e , by he inal e alua ion checkpoin , in e ac ions
exhibi ed signi ican e olu ion, shi ing owa ds mo e e ined solu ion de elopmen .
S uden s e isi ed ea lie con e sa ions o de elop inc easingly cohe en and obus
dialogues, e ec i ely employing i e a i e app oaches o p oblem-sol ing. This shi
was e iden h ough deepe i e a i e dialogues and mo e s a egic ollow-up
ques ioning wi hin he con ex o he design- hinking me hodology, p og essi ely
conside ing use needs, p elimina y p o o ypes, and, occasionally, ounda ional
business model elemen s. Ne e heless, u he enhancemen s in comp ehensi e
eal-wo ld easibili y assessmen s and de ailed ma ke analyses emain possible.
C i ical hinking imp o ed conside ably ac oss mos eams, demons a ed by hei
g owing capabili y o c i ically assess and challenge he ou pu s p o ided by
Cha GPT. Teams F and A exhibi ed excep ional pe o mance by sys ema ically
explo ing mul iple al e na i e scena ios, in eg a ing ex e nal e e ences, and explici ly
add essing p ac ical limi a ions o AI-gene a ed con en .
Addi ionally, signi ican ad ancemen s we e obse ed in communica ion cla i y and
o e all eam engagemen . Teams E and F consis en ly deli e ed in e ac ions
cha ac e ized by clea , con ex ually ich p omp s and e ec i e one. No ably, eams
ini ially exhibi ing lowe quali a i e in e ac ion quali y (speci ically Teams B and G)
demons a ed meaning ul p og ess by he conclusion o he s udy. Team G
ep esen ed an excep ional case, ini ially choosing a pe sonal paid accoun ins ead
o he p o ided Cha GPT accoun . Despi e his ini ial choice, hey signi ican ly
imp o ed hei in e ac ion quali y and oken usage by he inal e alua ion.
3.2 P esen a ion e alua ion
The p esen a ion assessmen s ca ego ized eams in o h ee dis inc g oups based
on hei demons a ed ou comes. The i s g oup, "De eloping eams" (A and E),
ul illed basic equi emen s bu lacked speci ici y ega ding a ge use s and in-dep h
explo a ion o easibili y o business conside a ions. Thei AI usage emained
minimal, mos ly limi ed o basic idea gene a ion and p oblem de ini ion.
The second g oup, "P o icien eams" (B, C, F, and G), p esen ed clea a icula ions
o hei challenges and solu ions, e ec i ely le e aging GenAI. Howe e ,
oppo uni ies emained o enhance echnical easibili y alida ion, ma ke esea ch
obus ness, and de ailed business modeling.The hi d ca ego y included a single
"Ad anced eam" (D), which demons a ed supe io dep h in p oblem-sol ing, s ong
AI in eg a ion, and ea ly alida ion o echnical easibili y and ma ke conside a ions.
This eam's p esen a ion indica ed a holis ic unde s anding o he design hinking
p ocess, subs an ially en iched by e ec i ely u ilizing Cha GPT.
Fu he analysis compa ing Cha GPT usage wi h he inal p esen a ion e alua ions,
as depic ed in Figu e 2, yields addi ional insigh s. The igu e illus a es a ela i ely
mino a iance (±10%) in usage e alua ion among eams compa ed o hei a e age.
Howe e , he p esen a ion e alua ions exhibi a no ably g ea e a iance, indica ing
mo e p onounced di e ences in eams' pe o mance.
Figu e 2. Cha GPT usage s P esen a ion e alua ions wi h de ia ions s he a e age g ade
o bo h Usage and P esen a ion e alua ions
No ably, "De eloping eams" (A and E) showed signi ican disc epancies be ween
hei Cha GPT usage sco es and hei p esen a ion e alua ions, wi h p esen a ions
g aded lowe compa ed o hei Cha GPT in e ac ions. Con e sely, he "Ad anced
eam" (D) ou pe o med in hei p esen a ion compa ed o hei Cha GPT usage
assessmen . These obse a ions sugges ha Cha GPT usage alone did no
equalize eam pe o mance, bu a he ele a ed he gene al s anda d o ou pu s,
ein o cing exis ing di e ences in luenced by in insic ac o s such as p io
knowledge, skills, mo i a ion, and eam dynamics.
Thus, while he in oduc ion o Cha GPT posi i ely impac ed o e all pe o mance and
enhanced engagemen and dep h in eam ou comes, i was no a decisi e ac o in
homogenizing he eams' ul ima e pe o mance le els. Ins ead, i ampli ied exis ing
pa e ns in eam dynamics, highligh ing he impo ance o a comp ehensi e,
human-cen e ed pedagogical app oach alongside echnological ools.
4 DISCUSSION AND CONCLUSIONS
The indings demons a e ha clea and in en ional inco po a ion o GenAI ools like
Cha GPT wi hin a s uc u ed challenge-based lea ning amewo k no ably imp o ed
s uden engagemen and ou comes. Ou esul s align wi h eme ging li e a u e,
showing enhancemen s pa icula ly in he p ocess o hypo hesis alida ion du ing he
need inding and idea ion phases, d i en by GenAI-gene a ed pe sonas o e ing
deepe insigh s compa ed o adi ional me hods.
Compa ed o i e p io cou se i e a ions, GenAI-gene a ed pe sonas enabled
s uden s o mo e e icien ly alida e hypo heses du ing need inding and idea ion,
p o iding deepe insigh s compa ed o adi ional su eys o in e iews. While
engaging eal s akeholde s emains ideal, p ac ical cou se cons ain s o en limi
such in e ac ions. Hence, GenAI p o ided a p agma ic al e na i e o op imize
alida ion and acili a e iche s uden ou comes wi hin limi ed cou se du a ion.
Fu he mo e, in oducing GenAI encou aged s uden s o explici ly decla e and e lec
upon hei AI usage, os e ing anspa ency, hones y, and academic in eg i y.
Con a y o ini ial conce ns abou educed e o due o GenAI use, ou esul s
showed ha eams employed hese ools no me ely o lessen hei wo kload bu o
achie e deepe explo a ion and mo e comp ehensi e ou comes. Consequen ly,
s uden p esen a ions exhibi ed inc eased con idence, clea e communica ion, and
highe -quali y con en .
Ou s udy sugges s ha he anspa en and explici in eg a ion o GenAI ools wi hin
challenge-based enginee ing cou ses signi ican ly bene i s bo h s uden lea ning and
acul y assessmen p ac ices. Con inued me hodological e inemen and u he
empi ical alida ion emain essen ial o ully ha ness hese p omising echnologies in
u u e educa ional con ex s.
4.1 Limi a ions and u u e wo k
Se e al limi a ions eme ged du ing his s udy, indica ing clea pa hways o u u e
me hodological imp o emen . Fi s , da a inconsis encies esul ed om he op ional
na u e o Cha GPT accoun use, leading some eams o p e e al e na i e,
non-moni o ed pla o ms. To add ess his, u u e implemen a ions should manda e
s anda dized GenAI pla o m usage o es ablish comp ehensi e logging sys ems
capable o cap u ing in e ac ions om all employed ools.
Second, acciden al dele ion o in e ac ion logs by some s uden eams ad e sely
a ec ed da a comple eness and accu acy. Fu u e i e a ions should include explici
guidelines combined wi h au oma ed backup mechanisms o ensu e obus da a
p ese a ion. Thi d, he GenAI-d i en assessmen me hodology could be u he
imp o ed h ough he adop ion o mo e igo ous e alua ion p ac ices, such as
majo i y- o ing schemes o agg ega ed sco ing ac oss mul iple GenAI-gene a ed
ub ic e alua ions (Ga cia Hue es e al., 2024).
Finally, addi ional esea ch employing longi udinal o compa a i e expe imen al
designs would signi ican ly enhance ou unde s anding o GenAI’s sus ained impac
on s uden s’ c i ical hinking skills, p oblem-sol ing capabili ies, and o e all
p o essional compe ence. Such s udies would p o ide aluable empi ical e idence o
ein o ce he pedagogical e icacy o GenAI in eg a ion in long- e m educa ional
con ex s.
5 ACKNOWLEDGEMENTS
This wo k has ecei ed suppo om he 2025 g an p og am o pa icipa ion in
con e ences and scien i ic publica ions in he ield o eaching inno a ion, p o ided
by he Ins i u e o Educa ion Sciences a UPC.
Gene a i e AI ools we e u ilized o language e inemen du ing he d a ing o his
manusc ip . Addi ionally, as explici ly men ioned wi hin he manusc ip , gene a i e AI
was employed o he usage assessmen o in e ac ion logs. The au ho s con i m ull
esponsibili y o he accu acy, o iginali y, and adhe ence o e hical and academic
s anda ds o all con en p esen ed.
