FinSight is a production-grade, 100% offline, local-first personal finance and expense tracker web application for people who want private, clear control of daily money. It records and imports income and expenses, presents financial activity and deterministic insights, and supports category budgets and savings goals. All currency is Indian Rupees (₹).
This document specifies the requested FinSight product, including its user-facing behavior, local data handling, page coverage, design system, and implementation stack.
FinSight provides a browser-based personal finance experience with a public entry page and private, locally persisted finance views. Its single active human persona records and imports transactions, reviews financial summaries and insights, manages budgets and savings goals, and controls local data backups. Parsing, scoring, persistence, and data visualization run on the device.
The current product includes the Dashboard, Transactions, SMS Import, Import Statement, Waste Detector, Analytics, Budgets, Savings Goals, and AI Insights. It also includes the Landing page as the public entry surface. The application does not provide cloud sync, backend APIs, external LLM inference, or transfer of user data off the device.
FinSight is a client-side application. The Landing page is anonymously reachable; the finance views require login according to the supplied page access contract. The source also requires zero backend API calls and all user data to remain on the device. Therefore, login is an access boundary specified by the page contract, but no server-backed authentication service or account-management workflow is defined here. The application must not send finance data to a backend. The implementation must reconcile the login access requirement with offline-only delivery without adding an unrequested remote identity service.
Transactions, budgets, savings goals, custom categories, and theme settings are stored in browser localStorage and mirrored in IndexedDB database finsight_db. SMS/UPI parsing, statement parsing, duplicate detection, deterministic financial health scoring, and rule-based coaching are performed locally. The PDF worker is bundled locally with Vite. No external CDN is used.
The current delivery includes the requested eight finance views plus Landing. The source does not define a future delivery horizon or additional future features.
HDFC, SBI, ICICI, Axis, Kotak, IDFC, Union Bank, IPPB, GPay, PhonePe, Paytm, BHIM, CRED, Navi Pay, and Cash. Preserve custom imported sources when editing.••••1234.papaparse, xlsx, and pdfjs-dist with a locally bundled Vite worker.₹1.2L and ₹24k.Custom badge.As a Personal finance tracker user, I should be able to open FinSight’s public entry page and proceed to the finance experience.
Provenance: required_inference for the Landing page; page access is explicit in the page contract.
Lifecycle and acceptance: The user opens Landing anonymously, sees the FinSight product context and entry action, and proceeds toward the accepted finance workflow. Finance destinations remain unavailable until the page contract’s login access condition is met. The implementation must not add a backend API or send finance data off-device. If the access condition cannot be satisfied under offline-only delivery, the app must communicate the limitation rather than imply a successful remote login.
As a Personal finance tracker user, I should see six months of realistic INR demo transactions, budgets, savings goals, and a sample custom category on first load.
Provenance: explicit for the seed-data commitment; required_inference for executing the first-load seed.
Lifecycle and acceptance: On first use, the app seeds six months of demo data so charts and metrics are populated. The data includes salary, freelance income, dividends, rent, Swiggy/Zomato/Darshini, Uber/Metro, BESCOM/APEPDCL/Airtel, Netflix/Spotify/Apple One, impulse buys, budgets, savings goals, and sample custom category Coffee & Chai \xe2\x98\x95. The user can review the resulting data on the relevant views. If seeding or persistence fails, the app reports the local failure and offers retry; it does not claim data was saved.
As a Personal finance tracker user, I should be able to review my financial position on the Dashboard.
Provenance: explicit in Planning Scope.
Lifecycle and acceptance: The user opens Dashboard and sees all-time balance; current-month income, expenses, savings, and remaining balance; six-month cash flow; financial health score; unnecessary-spending summary; top spending categories; total saved toward goals; and recent activity. Values reflect local data and use INR. If data is empty or unavailable, the page communicates that state and provides a local recovery path.
As a Personal finance tracker user, I should be able to add an expense or income transaction.
Provenance: explicit.
Lifecycle and acceptance: The user opens the add form from the top bar or Transactions, selects Expense or Income, enters a numeric rupee amount, selects a category, sets date and time, chooses a payment source, enters an optional note, and sets Recurring and Impulse buy status. Saving creates a locally persisted transaction and makes it available to Transactions and dependent summaries. Invalid input or persistence failure is shown without discarding the user’s values; the user can correct or retry. A successful save produces a status toast.
As a Personal finance tracker user, I should be able to edit an existing transaction.
Provenance: explicit.
Lifecycle and acceptance: The user selects a transaction, edits its accepted fields, and saves. The updated record is reflected in Transactions and dependent summaries. Existing custom imported payment sources remain available when editing. Validation or persistence failure preserves the edit state for correction or retry. A successful edit produces a status toast.
As a Personal finance tracker user, I should be able to delete a transaction.
Provenance: explicit in the requested transaction-change notifications and store behavior.
Lifecycle and acceptance: The user initiates deletion from the transaction workflow; the record is removed from local state and dependent summaries update. A successful deletion produces a status toast. If persistence fails, the app reports failure and retains or restores the prior record.
As a Personal finance tracker user, I should be able to use a built-in Other or Other Income category, or create and select a custom category.
Provenance: explicit.
Lifecycle and acceptance: In the transaction form, selecting Other or Other Income opens the nested Custom Category dialog. The user can choose the built-in category or create and select a custom category by entering its name and choosing an emoji and accent color. The new category is stored locally and selected for the transaction. Invalid or failed creation is reported and can be corrected or retried. Successful creation produces a status toast.
As a Personal finance tracker user, I should be able to use custom categories throughout FinSight.
Provenance: explicit.
Lifecycle and acceptance: Locally persisted custom categories are available in Budgets with a Custom badge, transaction entry, Transactions filters, Dashboard, Analytics, Waste Detector, and AI Insights. A category created or restored locally is reflected in those views. If local persistence fails, the app reports the failure and does not claim the category is available.
As a Personal finance tracker user, I should be able to paste and import bank SMS and copied UPI receipts.
Provenance: explicit; local parsing and storage are required_inference.
Lifecycle and acceptance: The user opens SMS Import from the top-bar quick action, Dashboard hero action, or page navigation; pastes one message, 20 or more mixed messages separated by blank lines, or multiline receipts from GPay, PhonePe, Paytm, BHIM, CRED, and Navi Pay; reviews parsed records; and imports them. Parsing and import occur locally. Unparseable records and duplicates are surfaced for review or correction. Successful imports update local transactions and produce a status toast; failures allow retry.
As a Personal finance tracker user, I should receive accurate supported dates and times from parsed messages.
Provenance: explicit.
Lifecycle and acceptance: The parser supports DD-MM-YYYY, DD/MM/YY, DD-MMM-YYYY (including 21-Nov-2026 and 21Nov26), YYYY-MM-DD, and written receipt dates such as 21 Nov 2026, 09:34 AM. It supports HH:MM, HH:MM:SS, 12-hour AM/PM, and 24-hour times, combines date and time into sortable ISO YYYY-MM-DDTHH:mm, and uses the current date/time only when the message omits a timestamp. The user can review the resulting timestamp before import. Invalid dates or times are reported rather than silently treated as exact.
As a Personal finance tracker user, I should not have account balances recorded as transaction amounts when importing messages.
Provenance: explicit.
Lifecycle and acceptance: Before amount extraction, the parser strips Avl Bal, Avail Bal, Avl Limit, Clear Bal, and Total Bal clauses. It parses Rs., Rs, INR, and ₹, including prefix ₹450.00 and suffix 450.00 INR. If no valid transaction amount remains, the record is flagged for review and is not silently imported as a valid amount.
As a Personal finance tracker user, I should receive the specified debit-before-credit classification for messages.
Provenance: explicit.
Lifecycle and acceptance: Debit takes precedence when both debit and credit wording appear. For example, “Your A/c X1234 is debited by Rs.320.00 and Swiggy UPI credited” is Expense (Debit). “credited by Rs.85,000”, “received Rs.500”, and “refund of Rs.299” are Income (Credit). The user can review the detected direction before import; ambiguous messages are flagged for review.
As a Personal finance tracker user, I should be able to identify payment source, merchant/payee, and category from supported messages.
Provenance: explicit.
Lifecycle and acceptance: The parser detects the exact supported sources HDFC, SBI, ICICI, Axis, Kotak, IDFC, Union Bank, IPPB, GPay, PhonePe, Paytm, BHIM, CRED, and Navi Pay; extracts merchant/payee names from UPI/.../..., to VPA ..., at <Merchant>, Paid to <Name>, or Info: ...; and masks sensitive account/card digits as ••••1234. It maps merchants and keywords to Food, Transport, Travel, or a matching custom category using the specified mappings. The user reviews the result before import; uncertain extraction is flagged rather than represented as certain.
As a Personal finance tracker user, I should be able to import statement files from my device.
Provenance: explicit; local parsing and storage are required_inference.
Lifecycle and acceptance: The user opens Import Statement, selects a CSV, XLSX, or PDF file, reviews locally parsed rows, and imports accepted records. CSV uses papaparse, XLSX uses xlsx, and PDF uses pdfjs-dist with coordinate-based row reconstruction and a local Vite-bundled worker. Duplicate detection is applied. Unsupported or malformed files produce an error and allow retry. Successful imports update local transactions and produce a status toast.
As a Personal finance tracker user, I should be able to review unnecessary spending and related suggestions.
Provenance: explicit in Planning Scope.
Lifecycle and acceptance: The user opens Waste Detector and reviews the locally derived unnecessary-spending summary and suggestions, including relevant custom categories. If no findings exist, the page shows a no-findings state. The user can continue to other finance views without an invented mandatory sequence.
As a Personal finance tracker user, I should be able to review detailed financial analytics.
Provenance: explicit in Planning Scope.
Lifecycle and acceptance: The user opens Analytics and reviews locally derived spending and financial analytics, including custom categories and six-month cash-flow context. Charts use INR and compact chart labels. Empty or unavailable data is communicated clearly.
As a Personal finance tracker user, I should be able to review and update category budgets.
Provenance: explicit in Planning Scope.
Lifecycle and acceptance: The user opens Budgets, reviews category budgets and locally calculated spending, and updates a budget. Custom categories are available with a Custom badge. Successful updates persist locally and produce a status toast; failures are reported and allow retry.
As a Personal finance tracker user, I should be able to review and update savings goals.
Provenance: explicit in Planning Scope.
Lifecycle and acceptance: The user opens Savings Goals, reviews goal progress, and updates a goal. Successful updates persist locally and produce a status toast; failures are reported and allow retry.
As a Personal finance tracker user, I should be able to review deterministic offline financial health scoring and rule-based coaching.
Provenance: explicit.
Lifecycle and acceptance: The user opens AI Insights or reviews the Dashboard health summary. The score and coaching are calculated deterministically from local data, with no external LLM call. The user sees the score and supporting insights; insufficient data is communicated as such.
As a Personal finance tracker user, I should be able to back up my local finance data.
Provenance: explicit.
Lifecycle and acceptance: From the slide-out drawer, the user selects Backup data. FinSight downloads a timestamped .json snapshot with { version: 2, exportedAt, transactions, budgets, goals, customCategories }. The user receives a status toast. If the download cannot be created, the app reports failure and allows retry.
As a Personal finance tracker user, I should be able to restore a local JSON backup.
Provenance: explicit.
Lifecycle and acceptance: From the drawer, the user selects Restore backup, chooses a .json file, and FinSight schema-validates and restores supported v1 or v2 backups locally. Invalid files are rejected with an actionable error and do not replace current data. Successful restoration produces a status toast.
As a Personal finance tracker user, I should be able to reset the application to demo data.
Provenance: explicit.
Lifecycle and acceptance: From the drawer, the user selects Reset demo data, confirms the action, and FinSight restores the six-month INR demo dataset, budgets, goals, and sample custom category. The user receives a status toast. Cancellation leaves current data unchanged; reset failure is reported without claiming success.
As a Personal finance tracker user, I should be able to switch between dark and light themes.
Provenance: explicit.
Lifecycle and acceptance: FinSight starts in dark mode with class="dark" on <html>. The user toggles dark/light mode in the top bar; the selected theme is stored locally and reflected across the application. Theme persistence failure is reported without implying the preference was saved.
explicit.Provenance: required_inference in Planning Scope. This is the sole accepted active human persona.
This user wants private, clear control of daily money and uses FinSight to record and import income and expenses, review financial position and spending, and manage budgets, savings goals, categories, and local backups. Their inputs include transaction amounts, categories, dates and times, payment sources, notes, recurring and impulse-buy choices, pasted bank or UPI messages, statement files, budget and goal updates, and backup files. They review parsed records and financial summaries, and decide whether to import or update data. Their work is distinct in that all finance activity remains on their device and insights are deterministic and offline. Success is observable when their local records, budgets, goals, and categories are reflected consistently across the relevant views and can be backed up or restored locally.
Other or Other Income, the nested Custom Category dialog offers the built-in category or creation of a custom category by name, emoji, and accent color.version, exportedAt, transactions, budgets, goals, and customCategories.Muse: Tobias van Schneider.
Headline: “Your money, in focus.”
FinSight uses a composed, precise, quietly confident dark editorial finance style. Financial data is the visual focus; do not use unrelated stock photography. The supplied images are style and visual-inspiration references for palette, typography mood, density, and component style, not instructions to copy their page structure.
#080E10; primary card #0F191C; elevated glass #152328 at 80% opacity; primary text #F8FAFC; secondary text #94A8AF; muted text #60737A; emerald primary #10B981; coral expense #F43F5E; amber warning and impulse #F59E0B; hairline border rgba(255, 255, 255, 0.07).#F4F7F6; card #FFFFFF; elevated surface #EBF1EF; hairline border rgba(15, 23, 42, 0.08); text #0F172A; primary emerald #059669; coral #DC2626; amber #D97706.-0.025em, line-height 1.18. Inter weights 400, 500, 600, and 700 for body copy, inputs, rows, chips, and buttons. Use tabular lining numerals for every currency amount, percentage, and chart axis.0.10em.backdrop-blur-md.The public Landing hero uses the accepted entry copy and the visual language of the Dashboard balance panel without presenting invented financial data. Place the oversized Playfair headline above a wide dark glass panel with a thin emerald edge. The panel’s visual focus is the product’s accepted balance concept, but it must not fabricate a user balance before local data is available. Keep “Add transaction” as the entry action and do not imply that a transaction can be saved before the access condition is met. Use the accepted “Paste Bank SMS / UPI” action only where the user can proceed to its accepted workflow. The composition uses a restrained emerald glow behind the panel, with no gradient-blob artwork or unrelated photography.
Interaction Model: Animated
Motion Tempo: restrained
Hero Dimensionality: flat
localStorage with an IndexedDB mirror named finsight_db, with automatic two-way synchronization.en-IN grouping such as ₹1,24,500, and use compact chart labels such as ₹1.2L and ₹24k.pdfjs-dist worker locally with Vite. Use zero external CDN.prefers-reduced-motion by disabling decorative movement and retaining a usable static arrangement.required_inference: Report local read/write, parsing, import, backup, and restore failures truthfully; do not claim success when an operation fails.motion/react).papaparse.xlsx.pdfjs-dist, coordinate-based row reconstruction, and a locally Vite-bundled worker.localStorage and IndexedDB database finsight_db.src/types/finance.ts, src/data/seedData.ts, src/utils/formatters.ts, src/utils/statementParser.ts, src/utils/insightsEngine.ts, src/hooks/useFinanceStore.ts, and modular src/components/... for TopBar, NavDrawer, HeroCard, Charts, Modals, and views.frontend (React application UI), backend (shared FastAPI process for application APIs and domain modules), and database (MariaDB for application-owned persistent data). These are provisioned services, not product requirements. FinSight’s explicit zero-backend-call and local-only data constraints mean the application must not use the backend or MariaDB for finance data or persistence. Preserve the provisioned services without connecting FinSight finance workflows to them.Food, Transport, Education, Rent, Bills, Shopping, Entertainment, Health, Travel, Subscriptions, and Other; and income Salary, Freelance, Dividends, and Other Income.₹.localStorage data.Impulse buy \xe2\x9a\xa1.
FinSight · Personal finance, kept local
Private, local-first finance tracking in Indian Rupees.
Transactions are available after you sign in. Offline access cannot complete that step.
Balance
No local data yet
No comments yet. Be the first!