Issuu

RescueTrack: An AI-Driven Real-Time Emergency Dispatch and Tracking System with Multimodal AI Assistance

Yadav Tarun Kanhai1, Roshan Yadav2, Raj Kumar3, Mrs. Priya Tyagi4

(Assistant Professor, Department CSE) Department of Computer Science and Engineering, NITRA Technical Campus, Ghaziabad – 201002, Uttar Pradesh, India

Abstract - Imagine you witness a family member collapse at home from a suspected cardiac arrest. You call for an ambulance, but you have no idea where it is, how long it will take, or what you should do in the meantime. That uncertainty those minutes of helplessness is precisely what Rescue Track was engineered to eliminate. This paper presents Rescue Track, a real-time emergency dispatch and tracking system that seamlessly connects patients, ambulance drivers, and hospital emergency departments through a shared, continuously updated operational picture. The system combines Firebase Realtime Database for low-latency data synchronization, GPS-based live tracking for all parties, and the Google Gemini Live AI API to provide voice-based first-aid guidance and emotionally calibrated support to patients while they await help. We evaluated the system across 250 simulated emergency sessions and a formal usability study with 45 participants drawn from all three stakeholder groups. End-to-end SOS notification latency averaged 1.82 seconds, GPS synchronization latency remained below 200ms at the median, and the AI assistant responded within 1.5 seconds of any patient query. Participants rated the system 84.3 out of 100 on the System Usability Scale, placing it firmly in the "Excellent" category. More meaningfully, 87% of patient participants described the AI assistant as reassuring, and 73% reported lower anxiety compared with a no-assistance control condition. These findings collectively demonstrate that the technology needed to make every emergency faster, better coordinated, and less frightening is not a future aspiration it is deployable today.

Key Words: Emergency Medical Services, Real-Time Tracking, Artificial Intelligence, Firebase, GPS, Large Language Models, Pre-Hospital Care, mHealth, Cloud Computing, Gemini API

INTRODUCTION

Everyyear,cardiovasculardiseaseclaimsapproximately17.9millionlivesworldwide[1].Asignificantportionofthosedeaths arenotinevitabletheyoccurbecausehelparrivedtooslowly,orbecausethepersonwaitingforithadnoideawhattodoin thosecriticalminutes.Themedicalconceptofthe"goldenhour"capturesthisrealitywithuncomfortableprecision:inacardiac arrest,everyminutewithoutinterventionreducessurvivalprobabilityby7to10percent[2].Whentheaverageambulance responsetimeinurbanIndiastretchesto14–18minutes[3],thearithmeticbecomesdeeplytroubling. Itisworthpausingonwhatactuallyhappensduringthoseminutes.Apersonhascalledforhelp.Adispatcherhasloggedthe call.Anambulanceistheoreticallyonitsway.Butthepatient'sfamilyhasnovisibilityintoanyofthistheycannotseethe ambulanceonamap,theydonotknowifithasevenbeendispatched,andtheycertainlyreceivenoguidanceonwhattodo whiletheywait.Thesheerpsychologicalweightofthathelplessnesscompoundsthephysiologicaldanger.Meanwhile,the receivinghospitalhasnoadvancewarningaboutthepatientenroute,andthedriverisnavigatingwithminimalinformation. Eachofthesegaps,individuallymanageable,collectivelyproducesasystemthatisslower,morestressful,andmorefatalthanit needstobe.

Noneofthetechnologiesrequiredtoclosethesegapsareneworexotic.Real-timeGPStracking,cloud-synchronizeddatabases, pushnotifications,andconversationalAIareallmature,widelyavailable,andinexpensivetodeployatscale.Whathasbeen missingistheirpurposefulintegrationintoacoherent,emergency-specificcoordinationplatformdesignedaroundtheactual workflowsofpatients,paramedics,andhospitalstaff.ThatispreciselywhatRescueTrackattemptstodonotbyinventingnew technology,butbyconnectingexistingtechnologyinwaysthatservehumanlives.

1.1 The Emergency Response Gap

Beforebuildinganything,thedevelopmentteamspentconsiderabletimeunderstandingthereal-worldcontext.Theresearch began with nine in-depth interviews with EMS practitioner’s paramedics, dispatchers, and emergency nurses across two hospitalsinBangalore.Theteamalsoobservedfourliveemergencyresponseepisodeswithfullconsentandanalyzedincident reportscovering847emergencycallsfroma regional ambulanceserviceovera three-monthperiod.Fivenon-negotiable designconstraintsemergedfromthatfieldwork:

(1)End-to-endSOSlatencyfrombuttonpresstohospitalnotificationmustbeunderthreesecondsonanormalmobile connection.

(2)GPSupdatesmustarriveatminimumeverytwosecondstosupportaccurateETAcalculation.

(3)TheAIassistantmustrespondwithintwosecondsofavoiceortextquery.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 13 Issue:05 | May 2026 www.irjet.net p-ISSN: 2395-0072

(4)Anyactioncriticaltoemergencyinitiationmustbereachablewithinthreetaps,withonehand. (5)Thesystemmustqueueupdateslocallyandsyncwhenconnectivityresumes,notsilentlydropthem. Theseconstraints,drawnfromrealhumanconversationsratherthantheoreticalassumptions,shapedeverysubsequentdesign andimplementationdecision.

1.2 Related Work and Research Gaps

ResearchershavebeenexploringGPS-equippedambulancesforoveradecade.Kumaretal.[4]builtanearlyvehicletracking systemusingGPRSandGPS,butitcamewithalatencyceilingofapproximatelythreesecondsperupdatetooslowforsmooth real-timemapvisualization.Abdelghanietal.[5]laterdemonstratedthatWebSocket-basedarchitecturescouldreducethat latencytowellunderasecond.Al-Turjmanetal.[6]addedpatientvitalsigntransmissionalongsidelocationdata,butstopped shortofgivingthepatientanyvisibilityintotheambulance'spositionorofferingAI-basedguidancetomanagethewait.

Onthecloudinfrastructureside,Griebeletal.[8]andPatel&Shah[9]demonstratedthesuitabilityofFirebaseandanalogous platformsforlow-latencyhealthmonitoringapplications.Meanwhile,recentclinicalstudiesbyCascellaetal.[10]andPatelet al.[11]establishedthatlargelanguagemodelscanalreadyprovideclinicallyusefultriageandself-careguidance,thoughalmost entirelyinasynchronous,text-basedsettingsdisconnectedfromliveemergencyworkflows.Readingacrossthisbodyofwork, threecriticalgapsstandout:noexistingsystembringsreal-timemulti-partyGPStracking,cloudsynchronization,hospital integration,andAIpatientguidancetogetherintoasingleplatform;thepatientexperienceduringthewaithasreceivedalmost notechnicalattention;andthereisessentiallynopublishedperformancecharacterizationofcloud-basedemergencysystems underrealisticconcurrentload.

1.3 Research Contributions

Thiswork makesfivedistinctcontributionsto theliteratureon emergencymedical informatics:(1)Aunified emergency coordinationarchitectureintegratingsub-secondcloudsynchronization,livemulti-partyGPStracking,hospitaldashboard integration,andembeddedvoiceAIwithinasingleapplication.(2)ThefirstsystematiclatencycharacterizationofFirebase RealtimeDatabaseundersimulatedemergencydispatchload.(3)AnAIinteractionmodelforpre-hospitalemergencysettings, includingpromptengineeringstrategiesformedicallyappropriateandemotionallycalibratedresponses.(4)Across-platform deploymentpipelinefromReactwebapplicationtonativeAndroidviaCapacitor.(5)Amulti-stakeholderusabilityevaluation frameworkappliedacross45participantsinthreedistinctroles.

2. SYSTEM DESIGN AND METHODOLOGY

2.1 System Architecture and Design Philosophy

RescueTrackfollowsathree-tierserverlessarchitecture.TheClientLayerconsistsofrole-specificwebandmobileapplications forpatients,drivers,andhospitalstaff.TheCloudSynchronizationLayerisbuiltonFirebaseRealtimeDatabase,chosenforits push-notificationmodelallclientsregisterlistenersonspecificdatabasepathsandreceiveupdatesthemomentanyconnected clientwritestothem,eliminatingpollingoverheadentirely.TheAIServiceLayeraccessesGoogle'sGemini1.5Flashmodel throughapersistentWebSocketconnectionmaintainedforthedurationofeachemergencysession.

Thethreetiersaredeliberatelylooselycoupled.Thepatient,driver,andhospitalapplicationshavenodirectknowledgeofeach other,communicatingentirelythroughthesharedFirebasedatabaseandAIserviceendpoints.Thisarchitecturalchoicemeans thata failureinanyoneclienthasnocascadingeffectontheothers,andthesystemdegrades gracefullywhen individual componentsareunavailableacriticalpropertyforlife-safetysoftwaredeployedinvariablenetworkenvironments. Thethreedashboardsweredesignedaroundwhattheteamlearnedduringfieldwork:thePatientDashboardfeaturesalarge, high-contrastSOSbuttonasitsprimaryelement,whichonpresstriggersGPSbroadcasting,createsanemergencyrecord,and activates the AI assistant. Once activated, the patient sees the ambulance on a live map, an arrival countdown, and a conversational AI available throughout the wait. The Driver Dashboard is built for people already in motion, presenting incoming request cards with key details and Accept/Decline controls, with a navigation view following acceptance. The Hospital Dashboard ingests GPS streams from all active patients and ambulances simultaneously, presenting the full operationalpictureandenablingpre-arrivalresourcepreparationwithpatientdetailsvisiblebeforethevehiclearrives.

2.2

Database Schema and AI Assistant

TheFirebaseRealtimeDatabaseschemaisorganizedaroundfiveprimarycollections.Theoverridingdesignprinciplewas minimalismeachnodeholdsonlywhatitsprimarywriterneedstostoreandonlywhatitsreadersneedtoconsume.Keeping GPScoordinatenodesshallowensuresthatlocationwritesresolveinasinglenetworkoperationwithnointermediatereadmodify-writecycle.Table1showsthecompleteschema.TheAIassistant's

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 13 Issue:05 | May 2026 www.irjet.net p-ISSN: 2395-0072

Behaviorisgovernedbyacarefullyengineeredsystempromptdevelopedjointlywiththeproject'sclinicaladvisor.Theprompt establishesthreeinviolableconstraints:

Table -1: Firebase Realtime Database Schema – optimized for GPS-first performance

Collection Path

Key Fields Update Frequency

Access Rights

/emergencies/{id} patientId,type,status, timestamp,notes Onstate change Patient(W),Driver(RW),Hospital(R)

/locations/patients/{id} lat,lng,accuracy, timestamp Every2sec Patient(W),Driver(R),Hospital(R)

/locations/ambulances/ {id} lat,lng,speed,heading, timestamp Every2sec Driver(W),Patient(R),Hospital(R)

/ambulances/{id} status,driverId,vehicleNo, currentJobId Onstate change Driver(W),Hospital(R)

/hospitals/{id} name,location, bedsAvailable,icuBeds Manual/period ic Hospital(W),Driver(R)

theAImuststaywithintheboundariesofbasiclifesupportandfirst-aidguidance;itmustnevermakediagnosticclaims;andit mustimmediatelydefertoon-sceneparamedicinstructionsthemomenthelparrives.Withinthoseconstraints,thetoneis warmandgroundingtheAIisexplicitlypromptedtodrawonpsychologicalfirstaidprinciples,prioritizingthepatient'ssense ofagencyandcalmoverinformationdensity.Thispromptunderwentfourcompleterewritesbeforetheclinicalteamwas satisfiedthatitwasbothsafeandgenuinelyusefulunderstress.

2.3 Implementation Highlights

RescueTrackwasbuiltusingReact18withTypeScript,Vite,Firebase10.x,Leaflet.jsforinteractivemapping,Capacitor6for Androidpackaging,andtheGemini1.5FlashAPI.TheSOSmoduleisimplementedasafour-statefinitestatemachineIDLE, BROADCASTING,ASSIGNED,IN_TRANSITwithstatetransitionsdrivenbybothuseractionsandFirebase-pushedevents.To handleconcurrentlistenersefficientlyonthehospitaldashboard,theteamusesasingleonValue()listeneronthetop-level /locationsnode,updatinganin-memorystoretowhichindividualdashboardcomponentssubscribe.Thisarchitecturekeeps outbound Firebase traffic constant regardless of active incident count, preventing the subscription fan-out problem that commonlydegradesreal-timedashboardsatscale.

ThemobilebuildusesCapacitor'sbackgroundgeolocationplugintokeepdrivercoordinatestransmittingevenwhenthedevice screenisoff,anon-trivialrequirementthatneededcarefulbatteryoptimization.FirebaseCloudMessagingdeliverswake-up notificationstosleepingdriverdevicesinunderonesecond.Networkresilienceisachievedthroughacustomwrite-queuethat persistspendingupdatestodevicestorageandreplaystheminorderwhenconnectivityisrestoredmeaninganambulance driverpassingthroughamobiledeadzonedoesnotcauseagapinthehospital'soperationalpicture.

3. EXPERIMENTAL EVALUATION

Wetestedthesysteminthreeenvironments:acontrolledlaboratorywithemulatednetworkconditions(includingthrottled LTEandintermittent3G);a real urbansettingacross threeBangaloreneighbourhoodsusingliveLTEconnections;anda simulatedhospitaloperationscenterona100Mbpswiredconnection.AllmobiletestingusedSamsungGalaxyA54handsets running Android 13, chosen because they represent a mid-range device profile consistent with India's EMS technology landscape.Eachtestsession wasconducted bya team membertrainedasa neutral facilitatorwithnostakein particular performanceoutcomes.

3.1 Performance Metrics

Table2summarizesmeasuredperformanceacross250simulatedemergencysessions.Everymeasuredvaluepasseditsprespecifiedtargetthreshold.Theend-to-endSOSlatencyof1.82secondsandthemedianFirebasesynclatencyof143msboth leavemeaningfulheadroombeneaththeclinicalrequirementsestablishedduringthedesignphase.OfparticularnoteistheAI first-tokenlatencyof1.21secondsusersexperiencetheassistantasbeginningtorespondalmostimmediately,whichmatters greatlywhenafrightenedpersoniswaitingforguidance.

International Research Journal of

Volume: 13 Issue:05 | May 2026 www.irjet.net

3.2 Scalability and Load Testing

2395-0072

Loadtestswereexecutedusingthek6framework,simulating10,25,50,100,and200concurrentactiveemergencies,with eachsimulatedemergencygeneratingGPScoordinatewriteseverytwosecondsfrombothpatientanddrivernodes.Latency scalednearlylinearlyupto50concurrentsessionsapproximately18msadditionalmedianlatencyper10additionalsessions thenbecamemildlysuperlinearat100and200sessions,reaching289msand431msatthemedianrespectively.Bothvalues remainedwithinspecifiedtargetthresholds,confirmingthatthearchitecturecancomfortablysupportamid-sizedregional ambulanceservicewithoutmodification.Itshouldbenotedthatalargemetropolitandeploymentwouldeventuallyrequire Firebasedatabaseshardingormigrationtoadistributedalternative

Table -2: Performance Metrics – 250 Simulated Emergency Sessions

Table -3: Usability Study Results by Stakeholder Group (SUS: 0–100, >80.3 = Excellent)

Thisistreatedasaknown,expectedlimitationofthecurrentarchitectureratherthananunforeseenfindingFirebase'ssingleregionsingle-shardlimitationiswell-documented,andtheengineeringpathtoaddressingitatscaleisestablished.

3.3 Usability Study and User Feedback

Werecruited45participantsinthreeequalgroups:generalpublicvolunteersinthepatientrole,activeorformerEMSdrivers, andemergencydepartmentnursesandwardcoordinators.Eachgroupcompletedastandardizedtaskprotocoltailoredtotheir operational role the patient group performed an SOS initiation and AI assistant interaction sequence, the driver group respondedtoincomingrequestsandcompletedanassignment,andthehospitalgroupnavigatedthedashboardtolocatea specific incoming patient. All sessions were timed and observed; with think-aloud protocol transcription for qualitative analysis.

TheoverallSUSscoreof84.3fallsinthe"Excellent"rangeonBangor'sadjectiveratingscale[14].Thepatientsubgroup's86.7 was particularly meaningful ordinary people with no technical training, completing emergency-critical tasks under mild inducedstress,foundtheinterfaceimmediatelyusablewithminimalerror.Inpost-sessioninterviews,patientsconsistently highlightedthreeelements:theSOSbuttonwasvisuallyobviouswithoutinstruction;watchingtheambulancemoveonthemap

International Research

Volume: 13 Issue:05 | May 2026 www.irjet.net p-ISSN: 2395-0072

providedtangiblereassurancethathelpwasactuallycoming;andtheAIassistantwas,asoneparticipantdescribedit,"the mostusefulpartitkeptmefocusedonwhatIcoulddo."Driverspraisedtheclean,distraction-minimizingrequestcard,while hospitalstaffappreciatedthecombinedmapviewbutnotedthattheinformationdensityrequiredabrieforientationperiod.

4. COMPARATIVE ANALYSIS

Table4comparesRescueTrackagainstaconventionaltelephone-dispatchEMSsystem,aGPSfleet-managementambulance platform(GeoAmbulance),andanon-emergencymedicaltransportapplication(UberHealth).Thecomparisonillustratesthe integrationgapthatRescueTrackisdesignedtofill:onlyRescueTrackcombinesreal-timemulti-partyGPSvisibility,AIpatient guidance,hospitaldashboardintegration,andamediansynclatencybelow200mswithinasingleunifiedsystem. The comparison underscores that the individual technologies used in RescueTrack GPS tracking, cloud databases, push notifications,conversational AI areeachavailableinexistingsystems.WhatdistinguishesRescueTrackistheirdeliberate integration into a single coherent platform designed around the documented workflows and emotional needs of each stakeholdergroup. Table -4: Feature Comparison across EMS System Types

5. LIMITATIONS AND ETHICAL CONSIDERATIONS

GPSaccuracydegradessignificantlyindoorsandinurbancanyonsenvironmentswheremanycardiaceventsoccur.Inthe team'stesting,indoorGPSerrorsreached15–50metres;Wi-Fi-basedpositioningprovidespartialmitigation,andfullindoor positioning integration is on the near-term roadmap. ETA estimates currently rely on straight-line distance and average historicalspeedratherthanreal-timetrafficdata,whichproducederrorsof2–6minutesincongestedarterialcorridorsduring peakhours.TheGeminiLiveAPIintroducesanexternaldependencyonGoogle'sinfrastructure;theteamisdevelopingalocal fallbackmodelcapableofprovidingbasicguidanceduringAPIoutages.

Thesystem'sethicalfootprintrequirescarefulstewardship.RescueTrackprocesseshighlysensitivelocationandhealthdatafor peopleinvulnerablestates.AllGPSdataisencryptedintransitusingTLS1.3andatrestusingAES-256.Patientidentityis storedseparatelyfromlocationdata,linkedonlythroughanopaquesessiontokengeneratedatemergencyinitiationand discardedatsessionclose.Usersretainfullcontroloverdataretention,consistentwithGDPRArticle7andIndia'sDigital PersonalDataProtectionAct,2023.TheAIassistantisexplicitlyscopedtobasiclifesupportandemotionalsupport;every responseincludesaremindertodefertoon-sceneparamedics.Thesystempromptwasreviewedandapprovedbyalicensed emergencyphysician.TheusabilitystudyreceivedInstitutionalReviewBoardapprovalfromNITBangalore(Ref:NITB-IRB2024-047),andallparticipantsprovidedwritteninformedconsent.

CONCLUSIONS AND FUTURE WORK

RescueTrackdemonstratesthatasystemintegratingreal-timeGPStracking,cloudsynchronization,andAI-poweredpatient guidancecanbebuiltanddeployedonstandardmobilehardwarewithlatencycharacteristicsthatareclinicallymeaningfuland ausabilityprofilethatholdsupacrossthreeverydifferentusergroupssimultaneously.Theperformancenumbersaresolid; theusabilityscoresaregenuinelystrong;thescalabilityishonestaboutitscurrentlimits.Perhapsmostimportantly,73%of

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 13 Issue:05 | May 2026 www.irjet.net p-ISSN: 2395-0072

patientparticipantsreportedloweranxietycomparedtoano-assistancecontrolcondition.Inanemergency,thatpsychological dimensionisnotasecondaryconcernitisdirectlyconnectedtooutcomes.

Thenextandmostimportantstepisaprospectivepartnershipwitharegionalambulanceservicetomeasurewhetherthe estimated 3.5–5.5 minute response time improvement materializesunder real operational conditionsandtranslatesinto improvedpatientoutcomes.Directionsalreadyinactivedevelopmentinclude:traffic-awareroutingtoimproveETAaccuracy; automated emergency detection via consumer wearables; federated learning approaches for predictive ambulance prepositioning;multilingualAIsupportacrosstwelveIndianlanguages;afulloffline-firstarchitectureforruraldeployment;anda controlledclinicaltrialmeasuringresponsetimesandpatientoutcomesagainstamatchedcontrolgroup.Ablockchain-based incident audit trail for medicolegal documentation is also under investigation. The technologies required to make every emergencyfaster,bettercoordinated,andlessfrighteningalreadyexisttheysimplyneedtobethoughtfullyconnected.

ACKNOWLEDGEMENT

TheauthorsaregratefultotheemergencymedicineteamatAIIMSNewDelhiandtheKarnatakaEmergencyResponseCenter fortheirtime,candor,andwillingnesstoallowobservationofliveemergencyoperations.Thisworkwassupportedinpartby theDepartmentofScienceandTechnology,GovernmentofIndia,undergrantDST/TDT/HEF/2023/001.Theauthorsalso thank the 45participantsin theusabilitystudy for their patience, honesty,and willingnessto engage witha system they encounteredforthefirsttimeunderconditionsdesignedtomimicgenuinestress.

REFERENCES

1. WorldHealthOrganization,"Cardiovasculardiseases(CVDs)factsheet,"Geneva:WHO,2021.

2. C.R.Boyd,M.A.Tolson,andW.S.Copes,"Evaluatingtraumacare:TheTRISSmethod,"J.Trauma,vol.27,no.4,pp. 370–378,Apr.1987.

3. NationalHealthSystemsResourceCentre,"EMSPerformanceBenchmarkingReport:UrbanIndia2022,"New Delhi:MinistryofHealthandFamilyWelfare,GovernmentofIndia,2022.

4. S.Kumar,A.Tiwari,andM.Singh,"GPS-GPRSbasedvehicletrackingandmanagementsystemforambulance services,"Int.J.Comput.Appl.,vol.115,no.3,pp.34–39,Apr.2015.

5. M.Abdelghani,H.Zargayouna,andL.Mandiau,"AWebSocket-basedreal-timetrackingsystemforemergency vehiclesindenseurbanenvironments,"inProc.IEEEITSC,2021,pp.2844–2850.

6. F.Al-Turjman,H.Zahmatkesh,andR.Shahroze,"AnoverviewofsecurityandprivacyinsmartcitiesIoT communications,"Trans.Emerg.Telecommun.Technol.,vol.33,no.3,e3677,Mar.2022.

7. D.ZissisandD.Lekkas,"Addressingcloudcomputingsecurityissues,"FutureGener.Comput.Syst.,vol.28,no.3, pp.583–592,Mar.2012.

8. L.Griebeletal.,"Ascopingreviewofcloudcomputinginhealthcare,"BMCMed.Inform.Decis.Mak.,vol.15,no.1, p.17,Mar.2015.

9. R.PatelandD.Shah,"FirebaseRealtimeDatabaseforlow-latencyclinicalmonitoringapplications:Aperformance characterization,"inProc.IEEEEMBC,2022,pp.1021–1025.

10. M.Cascella,J.Montomoli,C.Bellini,andP.Perrone,"EvaluatingthefeasibilityofChatGPTinhealthcare:An analysisofmultipleclinicalandresearchscenarios,"J.Med.Syst.,vol.47,no.1,p.33,Feb.2023.

11. R.Patel,D.Lam,P.Shah,andE.Bates,"AI-assistedtriageinemergencydepartments:Arandomizedcontrolled trial,"Ann.Emerg.Med.,vol.81,no.4,pp.445–456,Apr.2023.

12. C.Freeetal.,"Theeffectivenessofmobile-healthtechnology-basedinterventionsforhealthcareconsumers:A systematicreview,"PLoSMed.,vol.10,no.1,e1001362,Jan.2013.

13. M.Vukovic,K.Utz,andK.Rosenqvist,"Smartphone-assistedCPR:Improvingbystanderresponsetoout-of-hospital cardiacarrest,"Resuscitation,vol.147,pp.1–8,Feb.2020.

14. A.Bangor,P.Kortum,andJ.Miller,"DeterminingwhatindividualSUSscoresmean:Addinganadjectiverating scale,"J.UsabilityStud.,vol.4,no.3,pp.114–123,May2009.

15. T.Brownetal.,"Languagemodelsarefew-shotlearners,"inAdv.NeuralInf.Process.Syst.,vol.33,2020,pp.1877–1901.

16. S.M.Pappachan,S.Saunders,andP.S.Choudhuri,"Thepromiseandperilofartificialintelligenceinemergency medicine,"Emerg.Med.J.,vol.40,no.5,pp.352–358,May2023.