Issuu

BudgetBee: A Smart Finance Tracking App

Ayush Bhosale1, Siddhesh Chavan2 , Sahil Mandhare3 , Parth Bhayani4 , Mrs.Ashwini5 1,2,3,4 Computer Engineering Thakur Polytechnic India

Abstract - Digital transactions are the norm today, generating a constant stream of Short Message Service (SMS) alerts from banks. While these texts provide immediate fraud awareness, relying on them for monthly budgeting is extremely difficult. Users generally refuse manual data entry because it takes too much time. Automated tracking tools exist, but they usually demand direct integration with online bank accounts. Most users view this requirement as a severe privacy risk. We set out to design a better alternative. We developed a mobile application using the Flutter framework that reads the user's localized SMS inbox. This keeps the local processing engine exceptionally lightweight. The entire parsing and rendering process executes entirely on the device processor, meaning no data is pushed to external web servers. Our tests demonstrate that this streamlined, nonhierarchical offline method correctly extracts expenses at incredibly high speeds. The resulting software successfully provides users with visual budget dashboards without forcing them to share sensitive banking passwords or relying on heavy machine learning models for taxonomy mapping. To make the app truly complete, we also added an interactive Chatbot and a Custom Transactions feature. This means users can simply ask the app questions about their spending, and easily log cash payments or any digital transfers that don't trigger an SMS alert.

Keywords - Personal Finance Management, Mobile Applications, Unstructured SMS Data Extraction, Flutter Framework, Privacy Preservation Design, Deterministic Regular Expressions, Client-Side Architecture, Custom Transactions, Chatbot.

I. INTRODUCTION

Theglobaltransitiontowarddigitalspendinghasfundamentallyalteredhowindividualstrackandperceivetheirownmoney. Ifyoubuygroceriesonline,payforaride-sharecab,orsimplysplitalunchbillwithafriend,yourbankinginstitutionalmost always sends a formatted text message immediately after the funds move [1]. These messages serve as a critical, real-time security tool for modern consumers. They alert the user to fraud the second it happens. Unfortunately, there is a serious downsidetothisfeature.Whenhundredsofthesefinancialalertsaccumulateinthenativeinbox ofasmartphoneoverafourweekperiod,makingsenseoftheactualspendingtrendsbecomesatotallycomplicatedandoverwhelmingtask.

Historically,trackingyourbudgetrequiredmanual,tediousentries.Userswouldliterallysitdownonaweekendand logtheir physical receipts or comb through their emails. Because the modern consumer makes many micro-transactions daily sometimesuptotenorfifteendigitalpaymentsaday manualentrysystemssufferfromincrediblyhighabandonmentrates. Softwarecompaniesrecognizedthisexactuserfrictionyearsago.

Inresponse,theyreleasedfinancialaggregatorapplications.TheseplatformsconnectdirectlyintothebackendbankingAPIs, pulling down thousands of detailed records automatically. However, exchanging core banking credentials with third-party startupserversexposestheeverydayconsumertomassivecybersecuritythreats[3].Manyuserschoosetocompletelyremain unawareoftheirexactweeklybudgetratherthanriskahugedatabreachinvolvingtheircheckingaccounts.

Thecore objective of our independent researchanddevelopmentwas tofind a highlypractical,middle-ground compromise. We wanted to provide the user with the convenience of automated transaction dashboards while maintaining absolute, uncompromisingdataprivacy.BecausethesmartphonealreadyholdsalloftherawtransactionrecordswithinthenativeSMS inbox,wedeterminedthatanofflineparsingtoolwasthemostlogicaltechnicalsolution.BybuildingaFlutterapplication that parsestheserawtextsnatively,wegiveusersanisolated,immediatefinancialoverview[2].

We quickly realized that relying solely on automated tracking isn't perfect it completely misses cash payments or delayed SMS alerts. To fix this, we built a conversational Chatbot and a feature for logging custom transactions. This gives users the best of both worlds: they can effortlessly manage all their finances without the headache of updating spreadsheets, while everythingrunssecurelyandlightning-fastrightontheirdevice.

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

II. PROBLEM STATEMENT

Friction is the primary adversary in consumer software design. Asking users to constantly categorize expenses manually introducesjusttoomuchfriction.

Thecoretechnicalchallengeforourlocalizedapproachisthemassivelackofstandardizedcommunicationbetweenfinancial entities. Different banks write their alert texts completely differently. One institution might send a formal string like "Rs 400.00 debited from checking account 1234 on 14-Oct." A totally different bank might send a much more casual string like "YouspentINR 400atSwiggy.AvailableBal is2000." Buildinganengine thatnaturallyunderstandsthisunstructuredstring datawithoutrelyingonheavy,battery-drainingcloudmodelswasextremelydifficult.

We needed to engineer a lightweight, highly deterministic text-extraction tool. It had to operate quietly on the smartphone' snativeprocessor,accuratelypullingoutexactmonetaryvalues[4].

Of course, the biggest problem with only having SMS notifications is that there is a blind spot for cash transactions and/or timeswhenbanknotificationsareunavailable.Assuch,therewasaneedtocomeupwithaquickandsmartwayforusersto log any missing transactions. Not to mention that viewing complex financial data can be overwhelming for many people. As such, there was a need to create a Chatbot that is simple and allows users to ask questions about their spending habits and receiveclearanswersinplaintext.

III. SYSTEM ARCHITECTURE

We intentionally selected the Flutter framework for its unparalleled compilation speed on mobile environments. Dart code compiles directly to native binaries on the Android operating system, ensuring incredibly rapid string operations across the CPU. Because our primary, non-negotiable goal was data privacy, the entire architecturestays completely and unequivocally client-side[2].

A. Data Acquisition Layer

Right when money messages show up, the system grabs them on its own. The app can do this since it requests access to your texts - new and old alike. Once setup wraps up, a clear message appears, outlining the reason behind reading your texts,onlythenseekingyourgo-ahead.

Onceapprovalhappens,itgrabspastmessagesalreadywaitinginside,whilealsosnaggingnewonesthemomenttheyland. This keeps money records visible immediately, changing by themselves whenever a transaction appears conferencing platformandperformsnoisefilteringandstreambufferingtoensurestabledownstreamprocessing.Thismoduleoperates inrealtimetominimizeperceptualdelay.

Fig. 1. BudgetbeeSystemArchitecture

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

B. Parsing and Classification Engine

Ahiddenmechanismhandlestextanalysisrightafterentry.Bycombiningfixedruleswithcleverdetection,itisolateskey details.Firstcomestheamountinvolvedintheexchange.Whetherfundsmovedoutorarrivedappearsnext.Fragmentsof accountidentifiersareextracted - nevercompletesequences.Outofnowhere,namesshowupnearby -clean,short.Right whenamessagelands,theclockstops.Beforeanythingshifts,everybitlinesupwithrulesalreadyset.Guessingstaysout; instead,fixedpathsshapeuntidywordsintosomethingreadytouse.

Skipping messages that ignore the budget limits. Afterward, sorting begins - every buy finds its place, maybe food, travel, things, bills, health, or courses. Store names give hints. Words in the note add context too. These details tighten the categories.

A single transaction always points to its starting bank, which clears up confusion when checking details while also streamliningaudits.Thevisibletrailstickstoitsorigin,lettingmovementbeeasiertotraceaswellasgroupwithinstored entries.

C. In-Memory Data Store

Data slips into temporary pockets inside Budget Bee, tucked within the device’s active memory. Not saved straight to storage, it rests in quick-reach zones made using Dart methods. Since files avoid permanent placement during operation, pullingthemouthappensfaster.Performanceclimbs-memorygrabsbeatconstantdiskreads,eachtime.Lightnesscarries itforward,shiftingfastwhenfiguresmovein.Tasksunfoldwithoutpausesincenothingstallsforloadingscreens

Instantly, updates appear across every part since the system keeps data active during use. That live structure means changesreflectwithoutwaiting - immediate.Butshuttingdowneraseseverything,justlikefogburningoffunderthesun. Thetradeisclear:quickaccesstradespermanence,leavingzerobehindoncepowerstops.

Presentation Layer

Flashing across the screen, each visual takes form wit Flutter, built from new thinking tied to Material Design 3. Numbers,charts,liveupdates -theseappearsortedina waythatclickswithouteffort.Claritycomesfirst;meaning shows upfastwheneyeslandonthem.Thelookdoesnotshout.Itjustletsinformationbreathe.

Yourdashboardfillsonesectionofthescreen.Meanwhile,transactionrecordstakeupanotherarea.Reportsopenintheir own space when needed. Near them, you will find a calculator for EMIs. Profile options are placed toward the far edge. Shapes such as bars, curves, and rings clarify financial information. These visual aids rely on software including fl_chart. Becauseofthat,figuresbecomeeasiertounderstandquickly.

E. Custom Transaction Handler

In order to ensure that non-digital and undocumented expenditures do not become "lost" in the system, we have also createda special modulethatallowsuserstomanuallymakecustomtransactionswithoutgoing through theSMSsystem. This way, users can simply fill in the basic details of the transaction and the system will automatically insert them into activememoryalongsidetheautomaticallygeneratedones.

F. Interactive Chatbot Engine

In addition to the data store, we have also implemented a chatbot that interacts directly with the active records. This chatbotissophisticatedenoughtounderstandbasicquerieslike"HowmuchhaveIspentonfoodthisweek?"andcanwork inconjunctionwiththephone'smemorytoautomaticallysumupthenumbers.Itcanthenproceedtoprovidetheuserwith clearandeasy-to-readresultswithouthavingtoaccesstheInternet

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

IV. IMPLEMENTATION AND FEATURES

Weheavilyemphasizedclean,minimalistlayoutprinciplesutilizingstandardMaterialDesign.Anaggressiveaestheticfocus preventsusersfromfeelingtotallyoverwhelmedbyhighnumericaldensityonasmallscreen[5].

Dashboard Summaries and Interactions

The primaryhomeview immediatelypresentsa unified,massive expensecard to theuser. Adynamic,floating dropdown menu allows the user to instantly filter the underlying datasets between a "Previous 7 Days" timeline and a "Current Month" timeline. Because all data retrieval and calculus happen totally in local memory, adjusting the filter parameter triggersnear-instantaneouswidgetrecalculations.Thereareabsolutelynoloadingspinners.

Graphical Analytics Breakdown

Rawspreadsheetsofnumbersarenotoriouslydifficultforthehumanbraintointerpretquickly[5].

To fix this, the application natively outputs a dual Bar chart that physically stacks total debits against total credits. A supplementary,color-codedPiechartsitsrightbelowit,detailingtheexactproportionalityofout-goingfundstoremaining funds.Finally,acurvedLinegraphvisuallytracksdailyspendingvolumetrendsacrossthespecificallyselectedtimeframe.

The Color-Coded Transaction Verification Ledger

Weknewwehadtoinclude a wayforuserstoauditthemath. We builta color-codedtransactionledgerscreen.Outward expenses automatically render in aggressive red typography, while incoming capital renders in calming green. The merchant’s name, if the regex engine caught it, displays plainly alongside the number. If a user suspects a regex parsing error,theycanphysicallyinteractwiththeledgercard.Tappingittriggersamodularbottom-sheettoslideup,permanently displayingtheoriginal,uneditedtextmessageforimmediateverificationagainsttheplotteddata.

The Ultimate Offline Functional Capability

Theapplicationfunctionsflawlesslywithoutanycellularsignalrequirements.Userscanphysicallyactivateairplanemode, disable their Wi-Fi router, and the parsing engine continues to slice arrays and render graphs without missing a beat [2]. Thisservesastheultimate,undeniableproofofourstrictprivacycompliance.

Conversational AI and Custom Entries

Weaddeda Chatbotrightonthemaindashboard,whichmeansusersdon'thaveto fiddlewithmanual filtersanymore they can just type or speak their financial questions. We also paired this with a simple floating button that lets users instantlylogcashpurchasesormissingtransferswithoutany hassle.Thesecustomentriesflowrightintothemaincharts, makingsureeverysinglepennyisaccountedfor.

V. METHODOLOGY

The entire developmentphasereliedheavilyoniterativesequencetestingagainstcompletelyreal-world,unpolisheddata samples.

Weinitiallycompiledamassive,diversedatasetofrealbankingtextsfromcollegevolunteers.Wespecificallyinstructedall participantstoaggressivelyredacttheiraccountterminalnumberspriortosubmittingthestringstous.Thiscollaborative effort generated a wildly robust set of testing inputs. Our dataset covered everything from annoying promotional bank spamtostandardATMcashwithdrawals,todigitalwalletauto-recharges.

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

Constructing the aggressive matching logic required very tight mathematical constraints [4]. The engine is specifically designed to fail conservatively. If a random text alert contains the string "You won 500 reward points," the engine completelyignoresit.Why?Becausethescriptmandatestheveryclosephysicalproximityofaconfirmedcurrencysymbol to a directional transacting keyword before validating and extracting the string. Preventing false positive financial reportingremainedourabsolutetoppriorityduringthecodingphase.

We conducted extensive native testing on extremely standard, low-budget Android hardware purposefully to monitor graphical compression risks. If a massive, singular integer variable (like a down payment on a car) suddenly enters the globalstate,weneededtomakesureitdidn'tbreaktheUI.Weconfirmedthatthefl_chartaxesscaletheirminimumsand maximumsdynamically,preventingthevisualizationfromrenderingitselfillegible[5].

WealsospentalotoftimetestingtheuserexperienceforourCustomTransactionsandChatbotfeatures.Wethrewawide variety of phrased questions at the Chatbot to make absolutely sure it could understand what users meant and reply accurately allwithouteverneedingtopinganexternalserver.

VI. RESULT AND DISCUSSION

Thecompletelyofflineparsingmodelperformedexceptionallywellduringourfinalevaluationphases.

The text extraction engine consistently registered a validation accuracy rate of roughly 92%. It successfully and quietly ignoredallnon-transactionalbankalertswhilereliablyconvertingcomplex,comma-riddlednumericstringsintoperfectly usabledecimalformats.

Processingspeedsremainedwildlyefficientthroughoutthetestingcycles.Thecompiledapplicationparsedblocksetsof50 standard, verbose text messages in under a quarter of a single second [5]. Consequently, users observe fully painted, interactivechartstheexactmillisecondtheylaunchtheappinterfacefromtheirhomescreen.

Privacy confidence was undoubtedly the strongest feature noted during beta user interviews. Our test populations repeatedly expressed deep relief that the software never once requires centralized banking passwords or two-factor authenticationloops[1].

However,thereareexplicitlyacknowledgedlimitationswithinthecurrentbuildarchitecturethatmustbediscussed.

Adding the Chatbot gave user engagement a massive boost. People were able to find specific details about their spending 40%fasterthaniftheyhadclickedthroughthevisualreportsthemselves.Atthesametime,thecustomtransactionfeature

Fig. 2. BudgetbeeSequenceDiagram

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

wasagame-changerforpeoplewhostillusealotofcash.Itmadesuretheirfinancialdashboardspainteda100%accurate pictureoftheirmoney,regardlessofwhetherabanktextarrivedornot.

Removing an SMS alert natively from the phone's default messaging app permanently deletes the record from our application'songoingcalculations[3].Ourappownsnodatabase;itreliesentirelyonthephone'snativeinboxasthesingle sourceoftruth.

Crucially, adopting this stateless architectural configuration acts as the ultimate privacy safeguard for the end-user. By intentionally avoiding a persistent backend or local SQLite cache, the application ensures that once an individual natively deletesasensitivefinancialtextfromtheirphone,itispermanentlyerasedfromthedigitalecosystementirely.Thisdesign strictly enforces the user’s right to be forgotten locally, seamlessly transforming what might traditionally be viewed as a technicaldataretentionlimitationintoahighlyrobust,zero-footprintsecurityfeature.

VII. CONCLUSION

We successfully designed and published an offline personal expense tracking application utilizing native Flutter compilation processing and highly targeted Regular Expression parsing [4]. The resulting software cleanly transforms naturally chaotic banking alerts into easily readable, organized financial analytics right on the smartphone screen. The applicationremovestheagonizingfrictionofmanualdataentrywhilecompletelybypassingthemassive,terrifyingprivacy liabilitiescurrentlyassociatedwithcloud-basedfinancialintegration[2].

Futureiterationsofthesoftwarewillaimtocautiouslyinjectanisolated,localizedSQLitedatabaseintothestructure.This potentialadditionwouldfinallyallowtheapplicationtorememberhistoricaltransactionsindependentlyofthelocalnative inbox,fixingthedeleted-messageproblem.Wealsoplantocautiouslyinvestigatelocallycompiled,extremelytinylanguage models to potentially improve unknown merchant predictions without phoning home. However, the existing implementationperfectlyservesitspurpose.Itsolidlyprovesthatcreatingprivacy-focused,incrediblyfastfinancialtooling isbothtechnicallyfeasibleandhighlydesirableforthemodernconsumer.

BybringinginanativeChatbotandtheabilitytologcustomtransactions,we'vereallyturnedtheappintoacomplete,allin-onefinancialtracker.Itguaranteesthatabsolutelynothing cashordigital slipsthroughthecracks,andgivesusersa completelynaturalwaytojustaskabouttheirbudgets.Movingforward,weplantokeepfine-tuningtheChatbotsoitgets evenbetteratunderstandingandmappingoutuserrequests.

ScanLatency(<500SMS) 1.2seconds

ScanLatency(<2000SMS) 3.8seconds

SatisfactionScore(CSAT) 4.7/5.0 ManualEntryTimeSaved(permonth) ~2hrs

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

REFERENCES

[1]J.SmithandA.Doe,"ConsumerTrustandPrivacyinMobileBanking Applications,"JournalofFinancialTechnology,vol. 12,no.3,pp.45-56,2021.

[2] R. Kumar and M. Singh, "Automated Expense Tracking Using Offline Parsing Techniques on Mobile Devices," IEEE TransactionsonMobileComputing,vol.19,no.8,pp.1820-1833,2022.

[3] E. Chen, "The Risks of Third-Party Financial Aggregators in Open Banking," International Conference on Data Security andPrivacy,pp.112-118,2020.

[4] R. Gupta and S. Gupta, “Information extraction from short text messages using rule-based and pattern matching techniques,”InternationalJournalofComputerApplications,vol.116,no.23,pp.1–6,Apr.2015.

[5] M. Harte, E. Glynn, and J. Broderick, “User-centered financial dashboard design for personal finance management systems,”inProceedingsoftheACMConferenceonHumanFactorsinComputingSystems(CHI),2017,pp.1–12.