FinSavvy – Your AI Finance Companion

International Journal of Research in Engineering & Science ISSN:(P) 2572-4274 (O) 2572-4304 Available online on http://rspublication.com/IJRES/IJRE.html volume 9 Number 6, 2025 DOI: 10.5281/zenodo.18055092 Original Article @2025 RS Publication, [email protected] 248 “FinSavvy – Your AI Finance Companion” Angad Varma, Bhargav Desai*, Ayush S, Darshan Kagadal and Mrs .Bhavya V Department of Computer Science & Engineering Dayananda Sagar Academy of Technology & Management Bengaluru, India *Corresponding Author Email: [email protected] I NTRODUCTION FinSavvy – Your AI Finance Companion is envisioned as an intelligent, mobile-first platform that unifies daily money management, long-term financial planning, and compliance guidance for individual users within a single application, with a special focus on the Indian retail segment. In this context, salaried and self-employed individuals International Journal of Research in Engineering & Science Available online on http://rspublication.com/IJRES/IJRE.html ISSN:(P) 2572-4274 (O) 2572-4304 ARTICLE INFO ABSTRACT ©2025 RS Publication Paper ID: IJRES694B8656D4CC9 Published: 2025-12-25 DOI: https://dx.doi.org/ 10.5281/zenodo.1805 5092 Page No: 248-252 FinSavvy works as your AI finance companion. It pulls together a mobile system for managing finances. The setup uses machine learning to give personal advice on things like insurance, taxes, loans, investments, and budgeting for everyday users. The front end runs on Kotlin and Jetpack Compose. It connects to a Fast API backend. Plus, it has encrypted storage right on the device. All this keeps analytics private. At the same time, it delivers a smooth and engaging experience for users. The recommendation engine relies on Random Forest methods. It can also tap into large language model APIs if needed. This lets it suggest the right insurance products, loan choices, and tax documents. It bases those on the users profile and what they aim to achieve. Then there is the financial health dashboard. It spots anomalies and shows net worth, debts, and any gaps in coverage. Everything updates in real time. For planning steps like EMI schedules or savings tied to goals, FinSavvy treats cash flow as a Markov Decision Process. It applies Deep Q Learning to come up with repayment and investment plans. These adjust as needed and weigh risks, available cash, and growth over time. The app sends proactive alerts. It offers clear explanations for renewals coming up, ways to save on taxes, and how close you are to your goals. In the end, FinSavvy stays active as an AI powered helper. It focuses on boosting financial knowledge, following rules, and building stronger economic stability for regular users in India. Keywords— FinSavvy, AI, Finance Companion Cite This Paper: Angad Varma, Bhargav Desai*, Ayush S, Darshan Kagadal and Mrs .Bhavya V. (2025). "FinSavvy – Your AI Finance Companion". INTERNATIONAL JOURNAL OF RESEARCH IN ENGINEERING & SCIENCE (IJRES), vol. 9, no. 6, 2025, pp. 248-252. DOI: https://dx.doi.org/10.5281/zenodo.18055092 International Journal of Research in Engineering & Science ISSN:(P) 2572-4274 (O) 2572-4304 Available online on http://rspublication.com/IJRES/IJRE.html volume 9 Number 6, 2025 DOI: 10.5281/zenodo.18055092 Original Article @2025 RS Publication, [email protected] 249 increasingly handle multiple income sources, complex tax regimes, diverse loan products, and a wide variety of insurance and investment instruments, while most existing tools concentrate on isolated functions such as expense tracking or static budgeting. Consequently, users depend on fragmented spreadsheets, banking apps, and generic calculators that lack personalization, cannot reason over long horizons, and offer limited decision support, often resulting in sub-optimal choices for insurance coverage, tax planning, and debt management. What this research aims to, instead, is the description of general AI finance topics, and it integrates a production-level, MVVM architecture Android client tightly integrated with a Fast API backend and Google Vision-driven document intelligence for Indian finance documents. The originality of this research contribution is found in (1) having all the sensitive analytics locally, along with the use of cloud OCR, (2) the workflow involving the steps of OCR, document classification, structured transactions, and ML models within the process of generating recommendations, and finally (3) the design pattern, component, and template developed. A. Problem Statement Recent breakthroughs in machine learning (ML), essentially AI-powered personal finance assistants, have proven that systems centered on data are capable of providing personalized insights to individuals by learning from certain behavior, risk tolerance, and financial objectives, rather than purely on expert-defined rules. Most existing solutions, on the other hand, tend to either concentrate on a very limited application (like only a budgeting assistant), or are largely cloud-focused, which leads to certain misgivings with respect to confidentiality, safety, and regulatory norms, especially when dealing with highly private financial, bank, insurance, and/or tax-related data. Other concurrently ongoing research on reinforcement learning (RL), a particular pillar of ML, on application to finance problems, such as rebalancing, debt, or Equated Monthly Installments, has proven highly promising but is hardly used on a consumer level. B. Proposed Solution: FinSavvy Hybrid Architecture of Machine Learning for India FinSavvy serves present limitations through holistic supervised learning, ensemble learning, and reinforcement learning architecture that is peculiarly crafted for Indian financial scenarios. Intelligent Recommendation Engine An LLM API-powered recommendation engine, built on Random Forest, recommends suitable insurance products, tax return strategies, and loan options by leveraging each user's profile, cash flows, and financial goals. Financial Health Dashboard A holistic financial health dashboard consolidates on-device data of income, expenses, wealth, debt, and existing policies to calculate net worth; indicate gaps in coverage; and trigger anomaly indicators, thus providing the user with a clear, instantaneous picture of their current financial position. RL‑Based Long‑Term Planning The EMI planning and goal‑based saving tasks are modelled as a Markov Decision Process ‒ or MDP, solved with Deep Q‑Learning and obtain return-maximizing, risk‑aware, *Client Layer*(Android App) * MVVM layer Viewer View (Jetpack Compose UI: Splash, Dashboard, Document Scanner) View Model (state management, call of Use Cases/Repositories) Domain Layer ================ International Journal of Research in Engineering & Science ISSN:(P) 2572-4274 (O) 2572-4304 Available online on http://rspublication.com/IJRES/IJRE.html volume 9 Number 6, 2025 DOI: 10.5281/zenodo.18055092 Original Article @2025 RS Publication, [email protected] 250 Use Cases: Scan Document, Sync Transactions, Generate Data layer Repositories: Transactions Repository Local storage: “Encrypted Room/Data Store” for storing user profiles, transactions, policies, and models Backend Layer, Fast API RESTful API endpoints for: /health, /transactions (CRUD), /documents/upload, / Controllers / routers call: Transaction Service (business rules and validation) Document Service (Google Vision API Client + Keyword Classifier) Analytics/ML Service (server-side inference if you choose to offload any models to your service). AI & Data Services OCR services offered by Google Vision (bills, policies, Form-16, loan statements). Optional LLM API (explanations and summaries of text). Model store - RF models, anomaly detector, DQN policies if deployed server-side). Data Stores =============== On-device encrypted DB: User profile, transactions, model features, ML outputs. Backend DB (if used): user-associated but de-identified transaction records, log entries, and metadata. liquidity sensitive investment and repayment strategies. M ETHODOLOGY 1. Reliable Data Acquisition Process Data related to profiles, income, expenses, assets, liabilities, and current policies, which is verified by the user, is processed through the mobile app and stored in encrypted memory. This data is pre-processed, normalized, transformed into feature vectors, which are then used for training models using the random forest technique, anomaly models, or reinforcement learning models. 2. Random Forest Recommender System with LLM Capabilities Concerning insurance, taxes, and loans, some features such as risk profile, cash flow stability, credit history, or horizon of objectives are fashioned into a training or testing mechanism for a recommendation engine using the Random Forest approach. Finally, the best products or approaches are submitted to an API of an LLM that creates an interpretation of the results in natural language. International Journal of Research in Engineering & Science ISSN:(P) 2572-4274 (O) 2572-4304 Available online on http://rspublication.com/IJRES/IJRE.html volume 9 Number 6, 2025 DOI: 10.5281/zenodo.18055092 Original Article @2025 RS Publication, [email protected] 251 Financial Health Dashboard & Anomaly Detection Moreover, Financial Health Dashboard provides collective data from the device for calculating the net worth, coverage ratio, debt-to-income ratio, and savings rate through engaging graphics. Anomaly models point out instances of under-insurance, suspicious spending, or bad debt practices, which are then displayed in the form of notifications and indicators within the dashboard. MDP-Based Deep Q-Learning for Long-Term EMI planning and savings for the long-term objectives are formulated as a Markov Decision Process (MDP), with states being encodings of the user’s context, actions corresponding to funding/repayment decisions, and rewards capturing risk, returns, and liquidity considerations. A Deep Q-Learning agent is then developed to learn from simulated cash flow paths, allowing for the recommendation of optimized EMI plans through “what-if” analysis. C ONCLUSION AND F UTURE W ORK Conclusion The proposed FinSavvy – AI Finance Companion demonstrates how a hybrid ML architecture that combines Random Forest– based recommendations, financial health analytics, and MDP-driven Deep Q-Learning can deliver integrated support for insurance selection, tax optimization, loan planning, and goal-based investing in the Indian context. By keeping data encrypted on-device, augmenting predictions with LLM-generated explanations, and surfacing insights through a unified dashboard and proactive alerts, FinSavvy has the potential to improve users’ financial literacy, planning discipline, and savings outcomes, while reducing reliance on fragmented tools and manual tracking. Differing from previously developed AI-based finance assistants that tend to concentrate on costs alone or exculpate themselves from all responsibilities on cloud infrastructure, FinSavvy embodies the concept of client-centric, encrypted processing where documents obtained from OCR are directly inputted into the Random Forest and Deep Q-Learning algorithms on the device itself. Future Work Real-World Deployment and User Studies It is Roll out the FinSavvy system to a pilot group (e.g., students, young professionals) and investigate via longitudinal research the quantitative effects for savings rate, budget compliance, and insurance sufficiency, using survey-based methods as in previous research on AI-powered finance apps. More Rich Data Integrations through AA & Open Banking Plug into India’s Account Aggregator (AA) ecosystem or other open banking APIs to enable the safe import of bank statements, mutual funds, or credit information, allowing for better cash flow models with robust consent and privacy features. International Journal of Research in Engineering & Science ISSN:(P) 2572-4274 (O) 2572-4304 Available online on http://rspublication.com/IJRES/IJRE.html volume 9 Number 6, 2025 DOI: 10.5281/zenodo.18055092 Original Article @2025 RS Publication, [email protected] 252 Personalized Explainability & Behavioural Nudges Add an LLM layer for hyper personalized explanations, what-if analysis, or behavior nudges such as commitment devices or savings challenges that could be informed by research-validated findings about the role of fintech in influencing personal finance behavior. Improved RL Technology and Stress-Testing Research into distributional RL, multi-objective RL, and scenario-based stress testing to ensure that EMI policies and investment policies are robust to income shocks, markets, or regulations, using recent advances in RL applications in finance. Regulatory-Compliancesand-FairnessEstablish formal procedures for governance of models, such as testing for fairness/bias over various income classes, preparing model cards, and aligning with updated RBI/SEBI regulations on using AI in the finance sector, so that FinSavvy is kept trustworthy and watertight. Aspect FinSavvy Capability Guidance scope Insurance, tax, loans, EMIs, and goal - based saving Intelligence layer Random Forest, anomaly detection, Deep Q - Learning Data handling Encrypted on - device storage and minimal cloud use User impact (expected) Better budgeting, higher savings, less manual effort REFERENCES [1] Personal Finance Assistant with AI-Powered Budgeting – IJCRT, 2024 https://www.ijcrt.org/papers/IJCRTBH02008.pdf [2] An AI-Powered Personal Finance Assistant – IJOES, 2025 https://ijoes.vidyapublications.com/paper/Vol37/2.pdf [3] AI Based Personal Finance Management System – IJERT, 2025 https://www.ijert.org/ai-based-personal-finance-management-system [4] AI-Driven Personal Finance Assistants – RJ Wave / IJEDR, 2025 https://rjwave.org/ijedr/papers/IJEDR2502168.pdf [5] Reinforcement Learning for Personal Finance Management – IJACECT, 2025 https://journals.mriindia.com/index.php/ijacect/article/view/532