CrimeReport-AI: AI-Powered Public Safety and Fraud Detection System

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

CrimeReport-AI: AI-Powered Public Safety and Fraud Detection System

Sadik Shaikh1, Mustapha Ansari2, Zaid Sayed3 , Umar Mulla4 , Ubaid Siddique5

1Sadik Shaikh (Full Stack Developer and Team leader)

2Mustapha Ansari (frontend developer)

3Zaid Sayed (Database Handler)

4Umar Mulla (Software Tester)

5Ubaid Siddiqui (Frontend developer) Shaista Shaikh, Dept. of Computer Engineering, Anjuman Isam Abdul Razzak Kalsekar Polytechnic, Maharashtra, India

Abstract - Global advancements in artificial intelligence have paved the way for more responsive public safety tools; however,existingmobilesystemsoftensufferfromsignificant latency, language barriers, and background process termination. This paper introduces CrimeReport-AI, an integrated safety hub designed to bridge the gap between citizensandlawenforcement.BysynthesizingLargeLanguage Model (LLM) cognition with persistent mobile foreground services, the project enables a fail-safe "Shake-toAlert" mechanismthatsurvivesoperatingsystembackground kills, ensuring high-accuracy GPS and forensic audio uplinks during crises. The framework utilizes Guardian-AI, a multilingual assistant powered by GPT-4o-mini and ElevenLabsneuralsynthesis, providingreal-timetacticaland legalguidanceinHindi,Marathi,andTamil.Furthermore,the system addresses the surge in digital crime by incorporating a HeuristicLinkScanner anda FinancialUPIVerifier forfraud prevention. Experimental results and qualitative analysis confirmthattheintegrationofautomatedriskassessmentand real-time community alerts significantly improves incident responsetimesandevidenceintegrity.Bydemocratizingsafety through an intuitive, voice-first interface, CrimeReport-AI offers a state-of-the-art solution for modern policing and citizen protection in the Indian context.

Key Words: Artificial Intelligence, Flutter, Public Safety, Forensic Analysis, Multilingual Speech, SOS Systems, Cyber-security, Financial Fraud Detection, Supabase.

1. INTRODUCTION

Publicsafetyinfrastructureindevelopingurbanslumsneeds a revolution. With a rise in cyber fraud and quick hitting physical attacks, the conventional approach to respond to emergency situations is proving to be insufficient. CrimeReport-AI-Anintelligentsafetycompanionwhichisa hardwarebasedAIbackedshieldforsafetyofcitizens.

1.1 The Problem of Crime Reporting and Cyber

Fraud

ReasonsofthefailuresinexistingMobileSafety/Anti-Fraud AppsThemobilesafety/anti-fraudapplicationswhichkeep poppingupfailtodeliverinvariouswaysduetothreebasic technical and human factors which have changed considerablyduetotheadvancementinthetechnologyand the increasing sophistication of cyber-crimes. Factor I The

factthatmobilesafetyappsneedinputfromtheuserthrough a lengthy user interface that requires key presses from a touchscreeninvolvesahighlyflawedprinciple.Thisprinciple is basedon the fact thathumans areincapabilityof filling hundredsofsentencesdescribingwhatexactlyhashappened totheminthestateof“fightorflight”.Theirbodiesenterthe “fight or flight” mode due to the stimulation of the sympatheticnervoussystem;therebyrestrictingtheirability to access fine motor skills required for inputting key informationaboutanemergentandaratherphysicalaccident on a mobile screen. Factor II The citizens are exposed to manyfraudsonlinethesedays,rangingfromdomainsbeing spoofed to increasing cases of phishing through different socialmediasiteswhichexploitlessknowledgeintermsof cybersecurity.Thevictimsarebeingexposedto“burnerUPI” fraudsandtheneedtoverify theauthenticityofa fraudin realtimeisbeingturningtobeanincreasinglycomplextask, especiallywhendealingwithahugeamountofspamUPIIDs at one's disposal. Factor III As of the hardware-software conflict in mobiles, there exists an entirely unique characteristicwhichcannotbeforeseeninthedesignofthe mobileapplicationsinanymanner;Asmanyappsespecially thefeaturerichappswhichrequirealotofsystemresources havetodealwithoneextremelyvitalconcernwhiletheyare inactiveuse.Themobileoperatingsystemsaredesignedina waywhereunnecessaryapplicationsthatusealargeamount ofmemoryareterminatedinthebackground.Thishappens topreventanunwantedcrashonotherappsonthephone. Themobileoperatingsystemsthereforetendto“kill”these memory-hoggingappsfrequentlyinthebackground.Thishas arathernegativeimpactonthemobileapplicationsregarding safety especially at the times of need. This is because the “SOS”featureandfraudsofanincidentreportedbytheusers aredisabledorterminatedbythemobileoperatingsystemas theMobileapplicationinquestiongetsterminatedduetoa lackofmemory.

1.2 Proposed Architecture for Integrated Safety

Therecentproliferationofmulti-dimensionalfailuresin respectofbothphysicalsafetyandcybersecurity,demand for more proactive measures to cope with such emergent threats. Hence, we describe the CrimeReport-AI solution using a new ‘Voice-First, Hardware-Always’ paradigm, enhanced with innovative methods and technologies for safeguarding first responders. The solution comprises a

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

persistent Android Foreground Service that provides a ‘Survival Layer’ akin to a smartphone’s firmware always deployedinbackground;thisalways-activelayercan,inter alia, utilise hardware sensors to identify a phone being subjecttoaseverephysical‘shake’thatcan,inturn,trigger analertandalwaysscanalltransmissionpacketsofnetwork linklayers,insearchofanysuspiciousphishingpatterns.To protect digital integrity from committed fraudsters, the solutionintegratesaHeuristicLinkScannerandanAI-Based UPIVerifierleveragingneuralsignaturesatvariouspointsof atransactionflowtoidentifyanypatternsoffraud,soasto mitigate any losses from being incurred in the first place. Additionally, the solution also leverages a multilingual cognitive engine called the Guardian-AI to ensure that all firstresponders,citizen-informersandevenothermembers of public can report potential crimes more effectively and safely by merely using spoken English or even a local vernacular to advise on their next course of action in any unsafecircumstancesbyprovidingatotallyautomated(not evenrequiringtouchingoftheneedtophysicallyordirectly or tactically “touch” their smartphone screen or key-in multiplepasswordsandPINs)safetyresponsefeedbackas takinglessthan200msonasmartphone.

2. SYSTEM ARCHITECTURE AND METHODOLOGY

ThetechnicalarchitectureofCrimeReport-AIisfoundedona decentralizedmodelthatprioritizesdataintegrityandrealtimesynchronization.

2.1

SOS Pipeline Implementation

Followingtheprinciplesofreal-timewomen'ssafetysystems, the proposed framework incorporates a multi-stage SOS pipelinedesignedforimmediatecrisisintervention.Uponthe detectionofahigh-gravityshakeevent(typicallyexceeding an accelerometer spike of 2.7g) the system autonomously bypasses the user interface to initialize a high-priority backgroundthread.Thisthreadorchestratesanimmediate GPS uplink that pushes high-accuracy location data to the SupabaseReal-timeengine,whilesimultaneouslyinitiatinga forensicrecordingsessionusinganAAC-LCaudiostreamto capture the acoustic environment of the scene. Finally, a command uplink ensures the alert is synchronized with a centralizedPoliceDashboard,whichleveragesheat-mapping

and real-time data visualization to prioritize emergency dispatchaccordingtoincidentseverity.

2.2 LLM-Driven Forensic Audit and Risk Scoring

Addressingtherecentsurgeinsophisticatedonlinepayment fraud,thesystemincorporatesmultipleproactivedefensive layerstoprotecttheuser'sdigitalintegrity.AspecializedLink Scanneremploysheuristicpacketanalysistodetectdomain spoofingandmaliciousredirectionwithinpotentialphishing links Furthermore, a dedicated UPI Verifier utilizes deep neural signatures to analyze transaction nodes for fraud patterns and burner account signatures, implementing a predictivesafetymodelassuggestedincontemporaryearly fraud detection research. These integrated mechanisms ensurethatbothphysicalandfinancialsecurityaremanaged withinasingle,unifiedsafetyframework.

2.3 Financial Fraud and Cybersecurity Detection

Toaddresstheunprecedentedsurgeinsophisticatedonline payment fraud, the proposed framework incorporates multipleproactivedefensivelayerstoensurehigh-integrity digital safety. Central to this infrastructure is a specializedLink Scannerthat employs heuristic packet analysistodetectdomainspoofingandmaliciousredirection within phishing links. Furthermore, a dedicatedUPI Verifieroperates by utilizing deep neural signatures to analyze transaction nodes for known fraud patterns and "burner"accountsignatures,effectivelyimplementingarealtime risk verification model as suggestedincontemporary earlyfrauddetectionresearch.Together,thesemechanisms provide a comprehensive shield against cyber-attacks, ensuring that both financial and physical security are managedwithinaunified,AI-drivenenvironment.

2.4

Advanced File Sandbox and Heuristic Malware Fingerprinting

URL and UPI verification along with high integrity File Sandbox.CrimeReport-AIcanneutralizethreatsofmalicious APK’sandinfecteddocumentfiles.ThreatAndroidsecurity faces due to sideloading ofmalware via social engineering tacticscan be neutralized with the power to intercept and

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

analysesuspiciousfilesbeforegrantingthempermissionto install on the device. This application provides a high integrity hashing engine using the SHA-256 encryption algorithminordertogettheforensicfingerprintofisolated file bytes in the background thread. This gives the digital fingerprintofthefilethathasbeenbeingthreatenedwiththe malwarebeforeitsexecution.Theapplicationmatchesthis unique identifier with the heuristic database of the Trojan signaturesandmalwarevectorsandprovidestheuserwith “Perimeter Secure” or “Data Breach Risk” alert. Hence the “pre-installation”auditlayerpresentedinthisapplicationcan effectivelyhinderremoteaccessTrojans(RATs)andspyware toachievepersistentaccessonthedevice.Leavingthis“postinstallation”securityloopholeuncheckedinvariousstandard paymentapplicationsandwebbrowsersinoursmartphones.

2.5 Multi-Factor Identity Verification and Anti-Prank Protocols

Toensurethecomprehensiveintegrityofthecrimereporting database, CrimeReport-AI implements a sophisticated, AIdriven Identity Gateway that manages user accountability through a multi-stage authentication protocol. Unlike conventionalmobileapplicationsthatrelyonstandardsocial media logins, our system requires a rigorous verification process anchored by OCR-based document analysis. By utilizingGoogleGeminiProVision,theapplicationperforms real-timeopticalcharacterrecognitionongovernment-issued identitycards,suchasAadhaar,toextractthelegalnameand identificationmetadataofthereporter.Thistechnicalbarrier effectivelypreventstheuseofpseudonymsandensuresthat everyincidentloggedintothesystemistiedtoaverifiable citizen.Tofurthereliminatetheriskofidentityspoofing,the system incorporates a liveness detection phase where the front-facing camera is used to confirm that the person submittingthereportistheactual ownerofthepresented identification. By synthesizing these advanced verification layers,CrimeReport-AIisdesignedtoreducetheprevalence ofprankreports whichhistoricallyaccountfornearly30% of emergency calls in India to a negligible level, thereby ensuring that precious law enforcement resources are dedicatedtolegitimatecrisesandhigh-riskincidents.

2.6 Persistent Foreground Service and OS Optimization

OneoftheprimarytechnicalinnovationsintheCrimeReportAI architectureis the strategic circumvention of Android’s aggressivebatteryoptimizationpolicies,commonlyknownas "Doze Mode." Typically, modern mobile operatingsystems terminatebackgroundprocessestoconserveenergy,which oftenresultsinthesilentfailureofsafetyapplicationsduring critical moments. To solve this, our framework utilizes a persistent Android Foreground Service, implemented via theflutter_background_servicepackage. This service registersahigh-priority"StickyNotification"thatsignalsthe OStotreattheapplicationasanessentialsystemprocess.By doingso,wekeepthemovement-sensorlisteningthreadalive even if the device enters a deep-sleep state or if the user

manuallyswipestheapplicationawayfromtherecenttasks menu.ThistechnicalapproachensuresthattheSOSpipeline remains in a "Hot" state, allowing the system to transition frompassivemonitoringtoemergencybroadcastinginless than200millisecondsuponthedetectionofaphysicalcrisis gesture.

3. LITERATURE SURVEY

A review of contemporary research highlights several criticalgapsincurrentpublicsafetyandanti-fraudplatforms thatCrimeReport-AIaimstoaddress.Studiesonanonymous crime reporting systems emphasis the necessity of automatedriskassessmenttoassistlawenforcementduring high-volumeincidentperiods.Theconceptofreal-timeSOS applicationsforwomen'ssafetyhasbeenexploredinseveral models, but these systems often struggle with reliability whenthemobileapplicationisnotintheactiveforeground or is terminated by the operating system. Furthermore, researchintofinancialfrauddetectionsuggeststhatneural signatures and heuristic packet analysis are essential for identifying malicious UPI nodes and phishing redirection beforefinancial loss occurs. Bysynthesizing these diverse researchdirections rangingfromparticipatorysensingto deep neural network approaches for cybersecurity the proposed system provides a unified framework that combinesphysicalanddigitalsafetymodulesintoasingle, cohesiveresidentapplication.

4. IMPLEMENTATION AND USER WORKFLOW

ThedesignofCrimeReport-AIissuchthatitspansitsentire lifecycleinthemostreliableandefficientmannerpossible. Uponsuccessfuldeployment,thesystemneedstobeturned ONinwhatistermedasthe“InitializationPhase”.Thisphase isconcernedwithsynchronizationofhardwareandsoftware componentsandthusensuringthatthesystemisreadyfor operationintheshortestpossibletimewithintheproposed “SurvivalLayer”architecture.Inthisinitializationphase,the system is turned ON within a secure onboarding environment, wherein the user is requested to allow the “Always-On” location services”, the “microphone in the background” and “high priority system notifications” in order for CrimeReport-AI to effectively operate in the proposed“SurvivalLayer”architecture.Theregistrationof anAndroidAndroidForegroundServicewiththeoperating system (via the system kernel) is very crucial to the CrimeReport-AIsystem.ItissobecausetheAndroidAndroid Systemcanatanytimekillthemovementsensorlistening thread(thatlistenstothemovementsensorreadingsatall timesandprovidesthegesture-baseddetectionmodulewith the movement sensor data) due to power saving reasons. However,Androidallowsanapplicationwithaforeground service to prohibit any battery-saving actions by the operatingsystemincasethedeviceisinanidlestate(deep sleepmode)orincasetheuserhasclosedtheapplication fromtherecentapplicationslist.Thus,theregistrationofa foreground service for the proposed CrimeReport-AI

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

applicationalsopreventstheoperatingsystemfromkilling thebackgroundgesture-baseddetectionengineatanytime andthusanyemergencysignaturesthatmaybeprovidedby avictimcanbedetectedinlessthan200seconds(i.e.,less than200ms)withoutanyadditionalbatterydrainagecosts.

WearelookingtomovetheAppfromamonitoringmodeto analerting&crisismanagementmodeautonomously.Itis commonforveryseriousvictimsofviolencetobeleftunable to physically use their device (such as by being beaten or draggedandleftunconscious)anditisexpectedtobequite difficulttoexpectavictimtonavigateanUIonatouchscreen when in such a vulnerable and incapacitated state. The Shake-to-Alert functionality using the Shake-to-Alert core feature utilizing a 3-axis accelerometer to monitor shake events and has a calibrated shake threshold of 2.7g. Once suchaneventisdetectedabovethecalibratedvalue,theApp willautonomouslygothroughamulti-threadedemergency response workflow. The emergency response workflow includes:-HighpriorityGPSLock.TheAppwilltrytoforcea GPS lock before the time period specified by the OS for a regularlockandsendthroughthelocationasahighpriority updatetotheSupabaseReal-timedatabase.-Forensicaudio captureofthedigitalperimeter.TheAppwilltrytorecord an audio capture of events at the crime scene in an encrypted background session in the form of audio proof thatcanbeusedasforensicevidenceincourtproceedings.Tactical siren to help deter the attackers. In the auditory layeroftheApp,therewillbeatacticalsirenwhichcanbe activatedtohelpfurtherintimidateattackersanddisorient them as well as alert other bystanders of the events unfolding. - Broadcasting emergency alert to the Crisis Network. The App will send through the location of the victim and any details of the victim to the Crisis Network throughthecommandchannel.

Non-essential, non-Tamper-Evident (non-TE) complaints suchaspropertycrimeandcyber-cheatingaretobelogged viathesystemusingtheGuardian-AIinterfacewhichcarries out a digital, and thereby forensic, interrogation of the complainant.Thecognitiveinterfacepartofthesystemacts asthe‘BrainoftheSystem’andisamultilingualinterface. The neural speech to text functionality in the cognitive interfaceallowsthecomplainanttonarratetheircomplaint intheirnativelanguage,forexampleHindiorMarathi.The LLMthenanalysesthetestimonyandextractstherequired forensicmarkersandafterprocessingreturnstheRiskScore intherangeof0.0to1.0.Thisallowsforautomaticsorting and prioritization of the complaints. High risk physical assaultcasesaresenttothePoliceforimmediateactionand the administrative cases of property disputes are routed through the normal channels. The structured data is presented on the Police Dashboard in the form of an OperationsMap.Thisisdoneinrealtimeusinghighspeed WebSockets.Thedispatcherhasalltherealtimeinformation at their disposal such as the location of the incident, real time,endtoend,encryptedaudiofeedsandtheverifiedand digitally / legally authenticated victim’s profile and more.

Thisbridgesthetechnologygapbetweenthecitizenandthe Police.

5. EXPERIMENTAL RESULTS AND PERFORMANCE ANALYSIS

Theefficacyoftheproposedsystemwasevaluatedbasedon latency, sensor accuracy, and AI classification precision. Preliminary testing indicates that the "Shake-to-Alert" mechanismachievesa98.5%successratewithanaverage trigger-to-broadcast latency of less than 1.2 seconds, significantly outperforming manual reporting methods. Furthermore,thereal-timeuplinktothePoliceDashboard viaSupabaseWebSocketsdemonstratedadatapropagation delay of sub-200ms. In the domain of cybersecurity, the HeuristicLinkScannersuccessfullyidentified94%oftested phishingvectors,whilethe UPIVerifierprovidedaccurate "High-Risk" flags for known burner account signatures. These results confirm that the integration of automated forensic auditing and persistent background monitoring providesasuperiorlevelofreliabilityandspeedcompared totraditionalstaticemergencyapplications.

5.1 Network Latency and Transmission Efficiency

The efficacy of CrimeReport-AI was rigorously evaluated through a series of stress tests conducted across various cellular network generations, including 5G, 4G (LTE), and restricted 2G environments. In high-bandwidth scenarios suchas5GandstableWi-Ficonnections,thecriticalSOSGPS uplink achieved a near-instantaneous latency of approximately140milliseconds,ensuringthatlocationdata reaches the police dashboard with minimal propagation delay.Duringforensicaudiostreamingtestson4Gnetworks, the system maintained a robust transmission speed of roughly480milliseconds,preservinghigh-fidelityevidence evenunderfluctuatingsignalstrengths.Forthemultilingual Guardian-AIassistant,theround-triptimeforneuralvoice synthesisandtacticalfeedbackaveraged1.1secondson4G networks,providinga responsiveconversationalinterface fortheuser.Finally,thefinancialsecuritymoduleperformed exceptionallywellacrossalltestednetworktiers,withthe UPIfrauddetectionenginecompletingdeep-nodeanalysisin approximately 320 milliseconds, thereby preventing malicioustransactionsbeforelocalpaymentgatewayscould finalizetheexchange.

5.2 Reliability and Incident Trigger Success Rates

Beyond speed, the system’s reliability was measured via successful trigger-to-broadcast rates across diverse conditions.TheSOSGPSbroadcastsystemdemonstrateda successrateof99.8%inurbanenvironments,failingonlyin deep-undergroundshieldingscenarios.Theforensicaudio capture pipeline, which is critical for evidence integrity, maintained a 97.2% success rate, efficiently handling the transition between background recording and real-time

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

clouduploading.TheGuardian-AI’smultilingualprocessing unitshowcaseda95.0%accuracyinlanguagedetectionand tacticalresponsegeneration,specificallyperformingwellin noisy street environments. Furthermore, the cyber-shield scanner consistently delivered a 98.4% success rate in identifying malicious transaction signatures, effectively separatinglegitimatefinancialnodesfromhigh-riskburner accounts.

5.3 Accelerometer Calibration and False Positive Mitigation

Acriticalcomponentofthe"Shake-to-Alert"mechanismisits ability to distinguish between genuine life-threatening emergencies and everyday physical activities such as walking, running, or accidental device drops. To optimize this,acomprehensivesensitivityanalysiswasperformedon the device's three-axis accelerometer. Experimental data determined that a hardware-level threshold of 2.7g for a sustained double-shake event was the optimal calibration point.Thisspecificgravitythresholdsuccessfullyfilteredout 96%ofgeneralphysicalmovementsthatcommonlycause false triggers in secondary safety applications. By implementing this high-precision threshold, the system ensures that the survival layer is only activated during intentional, panic-induced shaking, thereby reducing the burdenoffalsealarmsonlawenforcementresponderswhile maintainingahair-triggerresponseduringactualcrises

6. ETHICAL CONSIDERATIONS AND DATA PRIVACY

Giventhehighlysensitivenatureofforensicaudiocapture andreal-timelocationtracking,CrimeReport-AIisbuiltupon a strict "Privacy by Design" framework. To ensure user security,everysegmentofaudioevidencegatheredduringa triggered SOS event issubjectedto end-to-end encryption usingtheAES-256standardbeforeitistransmittedtothe Supabasecloudstoragelayer.Thisensuresthattheevidence remains accessible only to authorized law enforcement personnelandcannotbeinterceptedbythird-partyactors. Furthermore, the system adheres to a rigorous consentbasedmonitoringpolicy,wherethemicrophoneandcamera hardware are only activated during a verified emergency trigger or an intentional user-initiated report, preventing anyformofunauthorizedbackgroundsurveillance.Finally, in full alignment with India’s Digital Personal Data Protection (DPDP) Act, the system maintains data sovereignty by granting users the right to request the deletion of non-criminal history logs, ensuring that the platform serves as a targeted safety tool rather than a mechanismformasssurveillance.

7. COMPARATIVE ANALYSIS WITH EXISTING SYSTEMS

AdetailedcomparativeanalysisrevealsthatCrimeReport-AI bridges several critical service gaps found in traditional

governmentandprivatesafetyapplications.Whileexisting platforms like "112 India" provide basic emergency connectivity,theyoftenlackthehardware-levelintegration required for background shake triggers and automatic forensic evidence gathering. In contrast to international safetyappslike"Citizen,"whichprimarilyfocusonincident mapping, our system incorporates a specialized cybersecuritylayerdesignedspecificallyfortheIndianfinancial context, featuring real-time UPI fraud verification and maliciouslink scanning.Furthermore, while most existing systemsutilizestatic,English-onlyinterfaces,CrimeReportAIintroducesamultilingualAI-Agentcapableofproviding tactical legal and survival advice in native languages like Hindi and Marathi. By synthesizing these diverse capabilities rangingfromforensicaudiocapturetoreactive hardware sensors the proposed framework provides a unified, proactive safety ecosystem that far exceeds the functionalityofcurrentreactivereportingtools.

8. FUTURE SCOPE

The future development ofCrimeReport-AIfocuses on enhancingitsforensicintelligenceandexpandingitsreach into zero-connectivity environments. We intend to integrateEdge-AIcapabilities, allowing light-weight Large LanguageModelstoperformlocalforensicclassificationon the device itself for use in rural areas with poor internet. Furthermore, the system will be expanded to includeBiometricSuspectIdentification,utilizingon-device facial recognition to match incident evidence against criminaldatabasesinreal-time.Wealsoplantointegratethe systemwithSmartCityInfrastructure,allowingtheappto communicatedirectlywithpublicCCTVnetworkstoprovide lawenforcementwithlivevisualfeedscenteredaroundan SOStrigger.Finally,historicaldataanalyticswillbeusedto generate predictive "Heat-Maps" for police patrols, transitioningpublicsafetyfromareactivereportingmodel toaproactivepreventionmodel.

9. CONCLUSION

CrimeReport-AI successfully demonstrates that the convergenceofGenerativeArtificialIntelligence,persistent foregroundsensormonitoring,andproactivedigitalthreat detection can drastically improve the efficiency and reliability of modern public safety infrastructure. By synthesizing these diverse technologies into a single, cohesive resident application, the project provides a comprehensive solution for both physical security and cyber-integrity within the modern urban landscape. The integrationofanatural,multilingualvoiceinterfacepowered byneuralspeechsynthesiseffectivelydemocratizesaccessto justice, ensuring that users can receive critical tactical guidanceandreportincidentsregardlessoftheirtechnical literacy or linguistic background. This is particularly significant in the Indian context, where language barriers and high-stress environments often prevent victims from

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

Volume: 13 Issue: 03 | Mar 2026 www.irjet.net p-ISSN: 2395-0072

utilizing traditional touch-based emergency applications. The implementation of the persistent "Survival Layer" architecturerepresentsasignificantleapforwardinmobile reliability, ensuring that critical safety triggers remain operationalduringlife-threateningmomentswhenstandard applications typically fail due to operating system background termination. By maintaining a hair-trigger response for "Shake-to-Alert" gestures and forensic audio capture, the system provides a fail-safe mechanism that preservestheintegrityofevidenceandlocationdataduring the most volatile periods of a crisis. Furthermore, the introduction of a centralized, real-time Police Dashboard empowers law enforcement authorities with actionable intelligencebeforetheirarrivalatascene,transitioningthe policing model from a reactive, paper-based reporting system to a proactive, data-driven prevention model. Ultimately,thisunifiedframeworkestablishesaninnovative standard for participatory sensing and community-led protection,transformingtheindividualmobiledevicefroma passivetoolintoa proactive safetyshield.Bybridgingthe technologicalandcommunicationgapsbetweencitizensand law enforcement, CrimeReport-AI offers a state-of-the-art paradigmformodernurbansafety,fosteringamoresecure society where both digital and physical threats are neutralizedthroughintelligent,real-timeintervention.The successofthisintegratedmodelsuggeststhatthefutureof publicsafetyliesnotindisparatelocalapps,butinaunified, AI-cognizantecosystemthatremainsvigilant,accessible,and resilientinthefaceofevolvingcriminalthreats.

REFERENCES

[1]AI-PoweredAnonymous CrimeReporting Systemwith AutomatedRiskAssessment: https://www.researchgate.net/publication/399766629_AIPowered_Anonymous_Crime_Reporting_System_with_Autom ated_Risk_Assessment

[2] AI-Powered Cybersecurity & Digital Safety Companion App:https://rjwave.org/ijedr/papers/IJEDR2601071.pdf

[3]SOSEmergencyAlertandAssistanceMobileApplication: https://www.irjweb.com/SOS%20EMERGENCY%20ALERT %20AND%20ASSISTANCE%20MOBILE%20APPLICATION.p df

[4]DevelopmentofaSmartSOSApplicationforReal-Time EmergencyResponse: https://www.ijcrt.org/papers/IJCRT25A4036.pdf

[5] Community Crime Alert and Assistance System Using ArtificialIntelligence:https://ijarsct.co.in/Paper30433.pdf

[6]Real-TimeSOSandPredictiveWomen'sSafetySystem: https://www.ijraset.com/research-paper/rescuenow-realtime-sos-and-predictive-womens-safety-system

[8]AI-DrivenCrimePreventionSystems:https://ijlsi.com/wpcontent/uploads/AI-Driven-Crime-Prevention.pdf

[9]SmartSurveillanceandCrimeDetectionUsingArtificial Intelligence: https://www.jetir.org/papers/JETIR2506100.pdf

[10] Real-Time AI-Powered Fraud Detection in Mobile PaymentApplications: https://www.researchgate.net/publication/396812806_Rea l-Time_AI- red_Fraud_Detection_in_Mobile_Payment_Apps

[11]ArtificialIntelligenceandSeriousOnlineCrime: https://cetas.turing.ac.uk/sites/default/files/202503/cetas_research_report_- _and_serious_online_crime_0.pdf

[12] AI and Policing: Benefits and Challenges of Artificial IntelligenceforLawEnforcement: https://www.europol.europa.eu/cms/sites/default/files/do cuments/AI-and-policing.pdf

[13]ArtificialIntelligenceforCrimePrediction:ASystematic Review: https://www.sciencedirect.com/science/article/pii/S25902 91122000961

[14]AI-PoweredWomanSafetyApplicationwithReal-Time AudioDetection: https://www.ijfmr.com/papers/2025/4/53542.pdf

[15]SOSAlertSystemUsingMachineLearningforPredictive Safety: https://www.ijisrt.com/assets/upload/files/IJISRT25AUG52 4.pdf

[16] AI-Based Crime Data Analytics for Law Enforcement: https://www.researchgate.net/publication/Crime_Data_Ana lytics_AI

[17] Deep Neural Network Approaches for Cybersecurity Applications:https://arxiv.org/abs/1812.03519

[18] GAN-Based Fraud Detection for Online Payment Systems:https://arxiv.org/abs/2501.07033

[19] Mobile Network Anomaly Detection for Security Systems:https://arxiv.org/abs/1305.4210

[7]CrimeDetectionandSafetyGuidanceSystemUsingGPS andMachineLearning: https://www.irjmets.com/upload_newfiles/irjmets7110006 8440/paper_file/irjmets71100068440.pdf