Au oma ed Assessmen En i onmen Fo
P og amming Assignmen s
Ishan Pandi a
Depa men o Compu e Enginee ing
S i Si asub amaniya Nada College o Enginee ing
Chennai, India
3122225001042
Sakaya Mil on R
Depa men o Compu e Enginee ing
S i Si asub amaniya Nada College o Enginee ing
Chennai, India
P o esso
Sa hak Da shan
Depa men o Compu e Enginee ing
S i Si asub amaniya Nada College o Enginee ing
Chennai, India
3122225001122
Abs ac —This e iew examines au oma ed assessmen sys ems
o p og amming educa ion, ocusing on he de elopmen and
implemen a ion o comp ehensi e e alua ion en i onmen s. The
e iew analyzes exis ing app oaches o au oma ed g ading, ma-
chine lea ning in eg a ion in code assessmen , and plagia ism
de ec ion mechanisms. Key indings indica e ha mode n au-
oma ed assessmen sys ems mus add ess mul iple e alua ion
c i e ia including co ec ness alida ion, pe o mance analysis,
code quali y assessmen , and academic in eg i y e i ica ion. The
e iew syn hesizes esea ch on con aine iza ion-based execu ion
en i onmen s, CI/CD pipeline in eg a ion, and AI-powe ed eed-
back gene a ion. Cu en implemen a ions demons a e success-
ul mul i-language suppo wi h s ong accu acy in complexi y
analysis, au oma ed pipeline deploymen using Kube ne es o -
ches a ion, and comp ehensi e da abase in eg a ion o scalable
s uden managemen . Fu u e di ec ions emphasize enhanced
AST-based pe o mance analysis, expanded AI in eg a ion o
nuanced e alua ion, and ex ended p og amming language sup-
po . This e iew con ibu es o unde s anding he e olu ion
o au oma ed p og amming assessmen and iden i ies c i ical
gaps equi ing u he esea ch in a eas o ad anced pa e n
ecogni ion, eal- ime code quali y e alua ion, and sophis ica ed
plagia ism de ec ion mechanisms.
Index Te ms—au oma ed assessmen , p og amming educa ion,
code quali y e alua ion, plagia ism de ec ion, con inuous in eg a-
ion, con aine iza ion, machine lea ning, complexi y analysis
I. INTRODUCTION
The exponen ial g ow h in compu e science educa ion has
c ea ed unp eceden ed challenges in p og amming assignmen
e alua ion. T adi ional manual assessmen me hods s uggle
wi h scalabili y, consis ency, and comp ehensi eness as en oll-
men in p og amming cou ses inc eases wo ldwide [1]. This
e iew examines he s a e-o - he-a in au oma ed assessmen
en i onmen s, analyzing how mode n sys ems add ess he
mul i ace ed equi emen s o p og amming educa ion.
P og amming assignmen e alua ion ex ends beyond simple
co ec ness e i ica ion o encompass pe o mance analysis,
code quali y assessmen , and academic in eg i y e i ica ion.
The eme gence o a i icial in elligence ools, pa icula ly la ge
language models capable o gene a ing code, has in oduced
new dimensions o he academic in eg i y challenge [8]. Simul-
aneously, he COVID-19 pandemic accele a ed adop ion o
emo e lea ning, c ea ing demand o obus au oma ed assess-
men sys ems capable o o line ope a ion while main aining
e alua ion igo [2].
This e iew sys ema ically examines exis ing au oma ed
assessmen app oaches, syn hesizes indings om ecen e-
sea ch in machine lea ning-based code e alua ion, and ana-
lyzes plagia ism de ec ion mechanisms speci ically designed
o p og amming assignmen s. The e iew c i ically e alua es
he s eng hs and limi a ions o cu en implemen a ions, high-
ligh ing a eas whe e exis ing solu ions demons a e e ec i e-
ness and iden i ying gaps equi ing u he esea ch.
The scope o his e iew encompasses au oma ed es ing
amewo ks, con aine iza ion echnologies o secu e code ex-
ecu ion using Docke and Kube ne es, con inuous in eg a ion
and deploymen (CI/CD) pipelines o assessmen au oma ion,
da abase sys ems o managing e alua ion da a, AI-powe ed
eedback gene a ion mechanisms, and mul i-c i e ia e alua ion
amewo ks in eg a ing co ec ness, pe o mance, quali y, and
o iginali y assessmen s [5].
The a ionale o his e iew s ems om he u gen need o
unde s and how au oma ed assessmen sys ems can e ec i ely
suppo p og amming educa ion a scale while main aining
educa ional alue and academic in eg i y. As a i icial in el-
ligence becomes inc easingly capable o gene a ing unc ional
code, educa ional ins i u ions equi e sophis ica ed ools ha
can dis inguish human-w i en code om machine-gene a ed
submissions while p o iding meaning ul eedback ha sup-
po s s uden lea ning [11].
II. LITERATURE SURVEY
A. Ea ly App oaches and Ou pu Compa ison Sys ems
The ea lies au oma ed p og amming assessmen sys ems
ocused p ima ily on ou pu compa ison, execu ing s uden
code agains p ede ined es cases and compa ing p og am
ou pu s wi h expec ed esul s. These sys ems p o ided basic
co ec ness alida ion bu lacked sophis ica ion in e alua ion
c i e ia and eedback quali y [4]. While e ec i e o in oduc-
o y p og amming cou ses wi h clea ly de ined inpu -ou pu
speci ica ions, ou pu compa ison sys ems p o ed inadequa e
o complex assignmen s equi ing subjec i e e alua ion o
code quali y, design pa e ns, and algo i hmic e iciency.
The limi a ions o ea ly sys ems mo i a ed esea ch in o
mo e comp ehensi e e alua ion amewo ks ha could assess
mul iple dimensions o p og amming compe ence simul a-
neously. This e olu ion e lec ed g owing ecogni ion ha
p og amming p o iciency encompasses no only he abili y o
p oduce co ec ou pu s bu also he capaci y o w i e e icien ,
main ainable, and well-documen ed code adhe ing o so wa e
enginee ing bes p ac ices.
B. Mul i-C i e ia E alua ion F amewo ks
No ak and Ke mek (2024) p o ide comp ehensi e analysis
o assessmen au oma ion o complex s uden p og amming
assignmen s, documen ing he e olu ion om simple ou pu
compa ison o sophis ica ed mul i-c i e ia e alua ion ame-
wo ks [9]. Thei esea ch iden i ies key ends including in-
eg a ion o s a ic code analysis ools o quali y assessmen ,
dynamic pe o mance p o iling o e iciency e alua ion, au-
oma ed es ing amewo ks suppo ing di e se p og amming
languages, and plagia ism de ec ion mechanisms adap ed o
code submissions.
The ansi ion o mul i-c i e ia e alua ion amewo ks ep-
esen s signi ican ad ancemen in au oma ed assessmen ca-
pabili ies. Mode n sys ems e alua e co ec ness h ough com-
p ehensi e es sui es co e ing edge cases and bounda y con-
di ions, assess pe o mance h ough algo i hmic complexi y
analysis and un ime p o iling, e alua e code quali y h ough
me ics including eadabili y, documen a ion, and adhe ence
o coding s anda ds, and e i y o iginali y h ough plagia ism
de ec ion and AI-gene a ed code iden i ica ion [3].
C. Sys ema ic Re iew o Cu en Tools
Messe e al. (2024) conduc ed sys ema ic e iew o au o-
ma ed g ading and eedback ools o p og amming educa ion,
analyzing ends in ool adop ion and iden i ying pa e ns
in implemen a ion app oaches [10]. Thei analysis e eals
g owing emphasis on p o iding meaning ul eedback beyond
simple co ec ness indica o s, in eg a ion o machine lea n-
ing echniques o unde s anding code seman ics, adop ion
o con aine iza ion echnologies o secu e code execu ion,
and implemen a ion o scalable a chi ec u es suppo ing la ge
s uden popula ions.
III. METHODOLOGY
A. La ge Language Models o P og amming E alua ion
Recen de elopmen s in a i icial in elligence ha e unda-
men ally ans o med au oma ed assessmen me hodologies.
La ge language models demons a e ema kable capabili ies
in unde s anding code seman ics, iden i ying logical e o s,
and gene a ing na u al language explana ions o p og amming
concep s [6]. This ad ancemen enables au oma ed assessmen
sys ems o p o ide eedback quali y app oaching human in-
s uc o s.
Mohammad e al. (2025) in oduced S epG ade, a no el ap-
p oach u ilizing con ex -awa e la ge language models o g ad-
ing p og amming assignmen s [11]. Thei esea ch demon-
s a es LLMs’ po en ial o unde s and nuanced aspec s o
code quali y adi ionally equi ing human expe ise. S ep-
G ade e alua es submissions h ough mul i-s age analysis in-
cluding syn ac ic co ec ness e i ica ion, seman ic analysis
o algo i hmic app oach, e alua ion o code e iciency and
op imiza ion, assessmen o eadabili y and documen a ion
quali y, and gene a ion o pe sonalized eedback add essing
speci ic weaknesses.
B. AI-Powe ed E o Analysis and Feedback
T adi ional au oma ed assessmen sys ems s uggle wi h
submissions con aining compila ion e o s, un ime excep-
ions, o incomple e implemen a ions. AI-powe ed e o analy-
sis add esses his challenge h ough sophis ica ed unde s and-
ing o code s uc u e and common p og amming mis akes [7].
The applica ion o la ge language models o e o analysis
enables sys ems o iden i y speci ic e o causes, explain why
e o s occu in accessible language, sugges conc e e ixes
add essing oo causes, and p o ide guidance o comple ing
pa ial implemen a ions.
IV. PROPOSED SYSTEM ARCHITECTURE
A. Assessmen Pipeline A chi ec u e
The comp ehensi e assessmen pipeline in eg a es mul iple
e alua ion modules o p o ide holis ic code analysis. Figu e
1 illus a es he comple e wo k low om submission o inal
g ading. The pipeline a chi ec u e o ches a es mul iple e al-
ua ion s ages:
Submission P ocessing: S uden s and ins uc o s submi
code h ough dedica ed in e aces, which ou e submissions o
he backend se e . Gi Lab se e in eg a ion enables e sion
con ol and collabo a i e de elopmen acking.
Compila ion and Valida ion: The compile s age pe o ms
ini ial syn ax alida ion and compila ion o compiled lan-
guages. Success ul compila ion p oceeds o co ec ness e alu-
a ion, while e o s igge he E o and Incomple e E alua o
using LLM analysis o cons uc i e eedback gene a ion.
Co ec ness E alua ion: Execu es comp ehensi e es sui es
compa ing p og am ou pu s agains expec ed esul s. Tes
cases co e no mal ope a ion, edge cases, and bounda y con-
di ions o ensu e obus co ec ness alida ion.
Cus om E alua ion: Enables ins uc o s o de ine special-
ized e alua ion c i e ia beyond s anda d co ec ness es ing,
suppo ing di e se assignmen equi emen s and lea ning ob-
jec i es.
Pe o mance Analysis: E alua es algo i hmic complexi y
using AST-based analysis and pa e n ma ching. De e mines
ime and space complexi y wi h con idence sco ing, p o iding
insigh s in o solu ion e iciency.
Fig. 1. Comp ehensi e assessmen pipeline showing e alua ion low h ough
compile alida ion, co ec ness es ing, cus om e alua ion, pe o mance anal-
ysis, plagia ism de ec ion, and code quali y assessmen , wi h LLM in eg a ion
o e o analysis and quali y e alua ion.
AI Plagia ism De ec ion: Employs sophis ica ed de ec ion
mechanisms o iden i y code simila i ies be ween submissions
and de ec AI-gene a ed code pa e ns, main aining academic
in eg i y.
Code Quali y E alua ion: LLM-based analysis assesses
eadabili y, documen a ion, coding s anda ds adhe ence, and
so wa e enginee ing bes p ac ices. Gene a es de ailed eed-
back wi h imp o emen sugges ions.
G ade Syn hesis: The g ade componen agg ega es esul s
om all e alua ion modules, applying con igu able weigh -
ing schemes o compu e inal sco es. Resul s pe sis in he
da abase o eco d keeping and p og ess acking.
B. Pe o mance Analysis A chi ec u e
The pe o mance analysis sys em employs a sophis ica ed
mul i-laye ed a chi ec u e designed o accu a e complex-
i y e alua ion ac oss mul iple p og amming languages. Fig-
u e 2 illus a es he comp ehensi e a chi ec u e in eg a ing
language-speci ic analyze s wi h co e se ices.
The a chi ec u e comp ises ou p ima y laye s:
In e ace Laye : P o ides mul iple access poin s including
CLI (Command Line In e ace) o di ec in oca ion, HTTP
API suppo ing Flask/Gunico n o scalable web access, and
con aine un ime enabling isola ed execu ion en i onmen s.
This lexibili y ensu es accessibili y ac oss di e en deploy-
men scena ios om de elopmen en i onmen s o p oduc ion
sys ems.
Co e Se ices Laye : The o ches a o componen manages
he analysis wo k low, ou ing eques s o app op ia e analyz-
e s based on language de ec ion. The analyze b idge se es
as an adap e laye in e acing wi h na i e language analyze s,
handling p ocess managemen and esul agg ega ion. Con ig-
u a ion managemen main ains analyze se ings and execu ion
pa ame e s.
Language Analyze s Laye : Con ains specialized analyze s
o each suppo ed language. The Py hon Analyze ope a es in-
p ocess using Abs ac Syn ax T ee (AST) pa sing o di ec
code analysis. The Ja a Analyze execu es as an ex e nal
JVM p ocess, u ilizing Ja aPa se o AST cons uc ion and
complexi y calcula ion. The C Analyze uns as a na i e bina y
compiled om C sou ce, employing libclang o sophis ica ed
AST a e sal. The Ja aSc ip Analyze ope a es h ough
Node.js, le e aging Babel pa se o mode n Ja aSc ip syn ax
suppo .
Suppo Se ices Laye : P o ides ounda ional capabili ies
including pa e n and ule eposi o ies con aining algo i hmic
pa e ns o sea ching, so ing, and common da a s uc u es;
complexi y calcula ion engines implemen ing ime and space
complexi y analysis algo i hms; da a models de ining esul
s uc u es and ep esen a ions; and u ili ies o ile de ec ion,
alida ion, and o ma ing.
C. Deploymen and Execu ion A chi ec u e
Figu e 3 p esen s he deploymen a chi ec u e u ilizing
Kube ne es o con aine o ches a ion, p o iding scalabili y
and eliabili y o p oduc ion en i onmen s.
The deploymen a chi ec u e implemen s se e al c i ical
componen s:
F on end Se ices: Use s in e ac h ough Monaco Edi o -
based in e aces, p o iding ich code edi ing capabili ies wi h
syn ax highligh ing and au ocomple ion. The ing ess con olle
manages ex e nal access and load balancing ac oss se ice
eplicas.
API Endpoin s: REST ul API se ices expose unc ionali y
o code submission, analysis eques s, and esul e ie al.
Buil on Flask wi h Gunico n wo ke s, he API laye handles
concu en eques s e icien ly while main aining esponsi e
pe o mance.
Caching and Queue Managemen : Redis p o ides high-
pe o mance caching o equen ly accessed da a and imple-
men s queue managemen o asynch onous analysis asks.
This a chi ec u e enables ho izon al scaling by dis ibu ing
wo k ac oss mul iple execu o nodes.
Da abase Pe sis ence: Pos g eSQL da abase main ains pe -
sis en s o age o use da a, submission his o y, analysis
esul s, and sys em me ics. Connec ion pooling op imizes
da abase access, while olume moun ing ensu es da a du a-
bili y ac oss pod es a s.
Execu o Nodes: Isola ed execu o pods pe o m ac ual code
analysis in secu e, esou ce-limi ed con aine s. This sepa a ion
Fig. 2. Pe o mance Analysis Sys em A chi ec u e showing in eg a ion o language-speci ic analyze s (Py hon, Ja a, C, Ja aSc ip ) wi h co e se ices including
o ches a o , analyze b idge, and suppo se ices o pa e n ecogni ion and complexi y calcula ion.
Fig. 3. Kube ne es-based deploymen a chi ec u e showing use in e ac ion low h ough on end in e aces o backend API endpoin s, wi h Redis caching,
Pos g eSQL da abase pe sis ence, and isola ed execu o nodes o secu e code execu ion.
ensu es ha po en ially malicious code canno comp omise
sys em in as uc u e while enabling pa allel p ocessing o
mul iple submissions.
D. Backend Implemen a ion De ails
The backend sys em implemen s se e al c i ical a chi ec-
u al pa e ns o eliabili y and scalabili y:
Da abase Laye : Pos g eSQL handles submission his-
o y s o age, use session managemen , language con igu a-
ions, and execu ion me ada a. Connec ion pooling op imizes
da abase access unde concu en load, while olume moun ing
ensu es da a pe sis ence ac oss sys em es a s.
Caching Laye : Redis manages execu ion queues, p o ides
as access o equen ly used da a, and handles empo a y
s o age o ongoing execu ions. This educes da abase load
and imp o es esponse imes o epea ed eques s.
API Se e : Exposes REST ul endpoin s o code submis-
sion wi h au hen ica ion and eques alida ion. Manages com-
munica ion be ween componen s and p o ides s a us upda es.
En i onmen a iable con igu a ion enables lexible deploy-
men ac oss di e en en i onmen s.
Wo ke Nodes: Execu e code in isola ed con aine s wi h
secu i y measu es and esou ce limi s. P ocess queued sub-
missions suppo ing mul iple p og amming languages while
managing execu ion imeou s and epo ing esul s.
Secu i y Fea u es: Con aine ized execu ion p o ides isola-
ion, implemen s esou ce limi a ions, and en o ces ne wo k
es ic ions. Sec e key-based au hen ica ion secu es API end-
poin s, while en i onmen a iable con igu a ion main ains
secu e c eden ial managemen .
V. PERFORMANCE EVALUATION AND RESULTS
A. Complexi y Analysis Accu acy
Comp ehensi e es ing o he pe o mance analysis module
ac oss 45 code samples demons a es s ong accu acy in algo-
i hmic complexi y de ec ion. Table I summa izes language-
speci ic pe o mance.
TABLE I
LANGUAGE-SPECIFIC COMPLEXITY ANALYSIS ACCURACY
Language Accu acy Sample Size
Py hon 100% 13 iles
Ja a 88.89% 9 iles
C 84.62% 13 iles
Ja aSc ip 70.00% 10 iles
O e all 86.67% 45 iles
Py hon achie es pe ec accu acy due o obus AST pa s-
ing capabili ies and comp ehensi e pa e n ma ching. Ja a
demons a es high accu acy wi h occasional misclassi ica ion
o complex ecu si e pa e ns. C analysis succeeds in mos
cases bu s uggles wi h sophis ica ed linea i hmic pa e ns
in ol ing nes ed loga i hmic ope a ions. Ja aSc ip p esen s
he g ea es challenge, wi h accu acy limi a ions s emming
om dynamic language ea u es and a ied coding s yles.
B. Sys em Pe o mance Cha ac e is ics
Pe o mance es ing e eals se e al c i ical sys em cha ac-
e is ics:
Analysis La ency: A e age complexi y analysis comple es
wi hin 0.8 seconds o ypical code submissions unde 500
lines. Analysis ime scales sub-linea ly wi h code size due o
e icien AST a e sal algo i hms.
Concu en P ocessing: The dis ibu ed a chi ec u e sup-
po s up o 50 concu en analysis eques s wi h Redis queue
managemen . Wo ke node isola ion p e en s esou ce con-
en ion be ween analyses.
Da abase Ope a ions: Que y execu ion imes emain un-
de 0.12 seconds o ypical e ie al ope a ions, suppo ing
esponsi e use in e aces. Connec ion pooling main ains pe -
o mance unde concu en load.
Con aine P o isioning: Docke con aine s a up a e ages
2.3 seconds, enabling apid scaling o handle submission
spikes du ing assignmen deadlines.
C. Da abase In eg a ion
Fig. 4. Da abase en ies showing pass/ ail esul s o mul iple es cases ac oss
34 submissions.
D. CI/CD pipeline
Fig. 5. Ac ual Gi Lab CI/CD pipeline execu ion o alida ion, es ing, and
g ading s ages.
Figu e 4 demons a es he da abase in eg a ion capabili ies
o he au oma ed assessmen sys em. The Pos g eSQL da abase
main ains comp ehensi e eco ds o submission e alua ions,
including de ailed es case esul s ac oss mul iple s uden
submissions. Each da abase en y cap u es submission me a-
da a, indi idual es case ou comes (pass/ ail s a us), execu ion
imes amps, and pe o mance me ics. This pe sis en s o age
enables ins uc o s o ack s uden p og ess o e ime, iden i y
common p oblem a eas ac oss he class, and p o ide a ge ed
eedback based on his o ical submission pa e ns. The da abase
a chi ec u e suppo s e icien que ying o gene a ing analy -
ics dashboa ds and p og ess epo s.
E. Limi a ions and A eas o Imp o emen
Despi e s ong o e all pe o mance, se e al limi a ions e-
qui e a en ion:
Linea i hmic Pa e n Recogni ion: The 75% accu acy o
O(nlog n)complexi y indica es need o enhanced pa e n
ma ching. Many linea i hmic algo i hms employ sophis ica ed
di ide-and-conque s a egies ha supe icially esemble sim-
ple pa e ns.
Dynamic Language Challenges: Ja aSc ip ’s 70% accu-
acy e lec s di icul ies in analyzing dynamic language ea-
u es. Dynamic yping, unc ional p og amming cons uc s,
and asynch onous ope a ions complica e s a ic analysis.
Recu si e Algo i hm Analysis: While simple ecu si e pa -
e ns achie e high accu acy, complex ecu si e algo i hms
wi h mul iple base cases o in e lea ed ecu si e calls p esen
analysis challenges.
Con idence Sco ing Calib a ion: Cu en con idence sco es
equi e e inemen o be e e lec ac ual accu acy. Some
high-con idence p edic ions p o e inco ec , while conse a-
i e sco ing unde es ima es accu acy o clea ly iden i iable
pa e ns.
F. Compa a i e Analysis O Exis ing Sys ems
Table II compa es majo au oma ed assessmen sys ems.
TABLE II
COMPARISON OF EXISTING SYSTEMS
Sys em Co ec ness Plagia ism Pe o mance
E alua ion De ec ion Analysis
CodeRunne Yes No No
MOSS No Yes No
Hacke Rank Yes Pa ial Yes
S epG ade Yes No No
P oposed AAE Yes Yes (AI) Yes
G. Tes Co e age
Fig. 6. PyTes execu ion ou pu and co e age epo o a sample s uden
submission.
H. G ading Co e age
Fig. 7. Fully au oma ed g ading esul summa y o a s uden submission.
VI. PLAGIARISM DETECTION AND ACADEMIC INTEGRITY
A. AI-Gene a ed Code De ec ion
The in eg a ion o De ec CodeGPT p o ides sophis ica ed
capabili ies o iden i ying machine-gene a ed code [8], [13].
The ool analyzes mul iple dimensions o code cha ac e is ics:
S ylis ic Consis ency: AI-gene a ed code o en exhibi s un-
usually consis en s yle h oughou , lacking he na u al a i-
a ions p esen in human-w i en code. Va iable naming con-
en ions, inden a ion pa e ns, and commen s yles main ain
uni o mi y exceeding ypical human a ia ion.
Commen Pa e ns: Machine-gene a ed commen s e-
quen ly e lec aining da a cha ac e is ics, using speci ic
ph asings and s uc u es common in documen a ion bu a e in
s uden code. Commen densi y and posi ioning ollow pa e ns
dis inguishable om human p ac ices.
Algo i hmic App oach: AI models end owa d ex book
implemen a ions o algo i hms, lacking he c ea i e a ia ions,
op imiza ions, and mino ine iciencies cha ac e is ic o human
p oblem-sol ing.
E o Pa e ns: Human code con ains cha ac e is ic e o
ypes e lec ing lea ning p og ession and common miscon-
cep ions. AI-gene a ed code, when con aining e o s, exhibi s
di e en e o pa e ns esul ing om model limi a ions a he
han concep ual misunde s andings.
B. T adi ional Simila i y De ec ion
Code simila i y analysis employs mul iple complemen a y
echniques:
Token-Based Compa ison: No malizes code by emo ing
whi espace, commen s, and o ma ing, hen compa es oken
sequences. This app oach de ec s supe icial modi ica ions like
a iable enaming and commen changes.
Abs ac Syn ax T ee Compa ison: Analyzes s uc u al sim-
ila i y by compa ing AST ep esen a ions. This echnique
iden i ies algo i hmically equi alen implemen a ions despi e
syn ac ic di e ences.
Con ol Flow Analysis: Examines p og am logic h ough
con ol low g aphs, de ec ing simila i ies in algo i hmic ap-
p oach e en when implemen a ion de ails di e .
VII. CONCLUSION
This e iew has examined he s a e-o - he-a in au oma ed
assessmen en i onmen s o p og amming educa ion, syn he-
sizing esea ch on e alua ion me hodologies, machine lea ning
in eg a ion, and plagia ism de ec ion mechanisms. The analy-
sis e eals signi ican p og ess in de eloping comp ehensi e
sys ems add essing mul iple e alua ion c i e ia while main-
aining scalabili y and eliabili y.
Key con ibu ions o cu en esea ch include demons a ion
o e ec i e mul i-c i e ia e alua ion amewo ks achie ing
s ong accu acy in complexi y analysis ac oss Py hon, Ja a,
C, and Ja aSc ip ; success ul applica ion o con aine iza ion
echnologies wi h Kube ne es o ches a ion o secu e, scal-
able code execu ion; implemen a ion o sophis ica ed CI/CD
pipelines enabling ep oducible e alua ion wo k lows; in eg a-
ion o la ge language models o e o analysis and code
quali y assessmen ; and de elopmen o specialized ools o
de ec ing AI-gene a ed code pa e ns.
The cu en s a e o knowledge indica es ha mode n au o-
ma ed assessmen sys ems ha e achie ed subs an ial unc ion-
ali y o comp ehensi e p og amming e alua ion. Success ul
implemen a ions demons a e eliable co ec ness alida ion
ac oss mul iple languages, e ec i e da abase in eg a ion sup-
po ing pe sis en da a managemen , unc ional o ches a ed
deploymen a chi ec u es, and ini ial AI in eg a ion o sophis-
ica ed eedback gene a ion.
C i ical gaps equi ing u he esea ch include enhanced
linea i hmic pa e n ecogni ion imp o ing accu acy a es,
Ja aSc ip analysis imp o emen s add essing dynamic lan-
guage ea u e challenges, ad anced ecu si e algo i hm analy-
sis o complex call pa e ns, con idence sco ing calib a ion
be e e lec ing ac ual p edic ion accu acy, and expanded
plagia ism de ec ion add essing e ol ing AI code gene a ion
capabili ies.
VIII. FUTURE DIRECTIONS
A. Uni ied E alua ion Pipeline
A uni ied e alua ion pipeline should in eg a e compile ali-
da ion, AI plagia ism de ec ion, co ec ness es ing, and LLM-
based e alua o s unde a single o ches a o . This cen alized
a chi ec u e would s eamline he assessmen wo k low, educe
p ocessing o e head, and enable coo dina ed e alua ion ac oss
all quali y dimensions.
B. LLM-Based In elligen Feedback Sys ems
Fu u e implemen a ions will le e age la ge language models
o p o ide sophis ica ed e o analysis and code quali y assess-
men . An in elligen e o and eedback sys em will in e p e
compila ion and un ime e o s, gene a ing na u al language
explana ions wi h s ep-by-s ep co ec ion guidance. The sys-
em will dis inguish be ween logical, syn ax, and concep ual
e o s o p ecise eedback. Addi ionally, an LLM-based code
quali y e alua o will assess eadabili y, modula i y, naming
con en ions, and e iciency while p o iding au oma ed e ac-
o ing sugges ions and s yle compliance checks.
C. ML-D i en Plagia ism De ec ion
Ad anced plagia ism de ec ion will employ deep lea ning
o unde s and code logic and seman ics beyond simple ex
simila i y. The sys em will pe o m seman ic equi alence
analysis o de ec plagia ism e en when code is es uc u ed o
ew i en, suppo c oss-language compa ison o iden i y unc-
ional equi alence ac oss di e en p og amming languages,
and p o ide clus e -based isualiza ion o code simila i y
g oups o ins uc o s.
D. G ade Syn hesize Module
A comp ehensi e g ade syn hesize will agg ega e all e alu-
a o ou pu s in o inal assessmen s. The module will implemen
weigh ed sco ing combining co ec ness, pe o mance, quali y,
and plagia ism me ics, gene a e LLM-assis ed explana ions o
g ading decisions, suppo adap i e g ading models adjus ing
weigh s based on di icul y and ins uc o con igu a ion, and
in eg a e wi h da abase sys ems o s uc u ed esul s o age
and dashboa d display.
E. Backend A chi ec u e Enhancemen s
Sys em scalabili y and eliabili y will imp o e h ough
a chi ec u al mode niza ion. Plans include adop ing mic ose -
ices a chi ec u e o sepa a e compile , e alua o , and AI
modules o independen scaling, implemen ing e en -d i en
wo k lows using message queues o asynch onous p ocessing,
deploying caching and load balancing wi h Redis and Nginx,
es ablishing secu e API ga eways wi h oken-based ou ing and
a e limi ing, and de eloping ins uc o con igu a ion po als
o cus om ub ic de ini ions.
F. Mul i-File and SQL Suppo
The sys em will ex end capabili ies o handle complex
assignmen s equi ing mul iple sou ce iles and da abase in-
e ac ions. Mul i- ile p ojec suppo will enable e alua ion o
modula p og amming assignmen s wi h p ope dependency
managemen and in e - ile ela ionship analysis. SQL suppo
will acili a e assessmen o da abase design, que y op imiza-
ion, and da a manipula ion asks. The e alua ion amewo k
will e i y que y co ec ness, assess pe o mance e iciency,
and analyze da abase schema design quali y.
REFERENCES
[1] C. S., S., C., B., & J., S. K. (2023). A Secu e and Scalable Sys em
o Online Code Execu ion and E alua ion using Con aine iza ion and
Kube ne es. Jou nal o Eme ging Technologies and Inno a i e Resea ch
(JETIR),10(2), c197-c202.
[2] Rau , S. S., Bamane, P. D., e al. (2022). A No el App oach o De ec ing
Malicious Ja aSc ip Code Using Machine Lea ning. In 2022 6 h
In e na ional Con e ence on Compu ing, Communica ion and Con ol
(ICCUBEA) (pp. 1-6). IEEE.
[3] Ca doso, M., de Cas o, A. V., Rocha, ´
A., Sil a, E., & Mendonc¸a, J.
(2020). Use o Au oma ic Code Assessmen Tools in he P og amming
Teaching P ocess. In 1s In e na ional Compu e P og amming Edu-
ca ion Con e ence (ICPEC 2020) (Vol. 81). Schloss Dags uhl-Leibniz-
Zen um ¨
u In o ma ik.
[4] Pa il, P. B., e al. (2023). Malicious Ja aSc ip De ec ion using Machine
Lea ning. In 2023 3 d In e na ional Con e ence on Pe asi e Compu ing
and Social Ne wo king (ICPCSN) (pp. 1126-1131). IEEE.
[5] Fang, Y., e al. (2019). A Machine Lea ning App oach o De ec
Malicious Ja aSc ip . In 2019 IEEE In e na ional Con e ence on Big
Da a (Big Da a) (pp. 4334-4339). IEEE.
[6] Rozi`
e e, B., Geh ing, J., Gloeckle, F., Soo la, S., Ga , I., Tan, X., ... &
Synnae e, G. (2023). Code Llama: Open Founda ion Models o Code.
a Xi p ep in a Xi :2308.12950.
[7] Mi chell, E., Zhou, Y., Khaza sky, A., Singh, M., Piech, C., & Finn, C.
(2023). De ec GPT: Ze o-Sho Machine-Gene a ed Tex De ec ion using
P obabili y Cu a u e. a Xi p ep in a Xi :2301.11305.
[8] Su, J., Zhuo, T. Y., Wang, D., & Nako , P. (2023). De ec LLM:
Le e aging Log Rank In o ma ion o Ze o-Sho De ec ion o Machine-
Gene a ed Tex . a Xi p ep in a Xi :2306.05540.
[9] M. No ak and D. Ke mek, “Assessmen Au oma ion o Complex S uden
P og amming Assignmen s,” Educa ion Sciences, 2024.
[10] M. Messe , N. C. C. B own, M. K¨
olling, and M. Shi, “Au oma ed
G ading and Feedback Tools o P og amming Educa ion: A Sys ema ic
Re iew,” ACM T ansac ions on Compu ing Educa ion (TOCE), 2024.
[11] A. Mohammad, K. Zami i Aza , and H. Kamali, “S epG ade: G ading
P og amming Assignmen s wi h Con ex -Awa e LLMs,” a Xi p ep in ,
2025.
[12] S. Bama, N. Mala, P. Ranjani, M. Saswa i, and A. T. Jeena, “Au oma ed
Compu e P og am E alua ion and P ojec s – Ou Expe iences,” 2024.
[13] Y. Shi, H. Zhang, C. Wan, and X. Gu, “Be ween Lines o Code:
Un a eling he Dis inc Pa e ns o Machine and Human P og amme s
(De ec CodeGPT),” in P oc. 47 h In . Con . on So wa e Enginee ing
(ICSE), 2025.
