Vani-Fi (वाणी-Fi) 🇮🇳

Vani-Fi is an AI-driven, vernacular-first financial intelligence ecosystem architected specifically for emerging digital banking users across Bharat (Tier 2, Tier 3, and Tier 4 towns).

Unlike conventional predatory fintech applications designed to maximize aggressive credit cross-selling and trap vulnerable users in high-interest debt cycles, Vani-Fi prioritizes financial well-being and knows when NOT to sell a loan. It pairs real-time cashflow intelligence with voice-driven conversational planning, dynamic safe-to-spend computation, and regulatory-grade RBI Key Fact Statements (KFS).

1. Predatory Credit Pushing: Most financial platforms nudge loans at every touchpoint, especially targeting financially stressed citizens with high-APR instant credit.
2. The "Account Balance Illusion": An account balance of ₹25,000 does not mean the user can spend ₹25,000. Rent, school fees, and electricity bills due in 10 days will bounce if funds are spent today.
3. The Linguistic & Literacy Barrier: Standard banking jargon (APR, compound interest, moratorium, penal charges) intimidates users who operate comfortably only in Indian languages like Hindi, Gujarati, Marathi, Tamil, or Bengali.
4. Opaque Lending Terms: Critical borrowing clauses, processing fees, and foreclosing penalties remain buried in multi-page English PDFs.

+-----------------------------------------------------------------------------------+
|                                  USER TOUCHPOINT                                  |
|          Indic Vernacular Voice AI (Hindi, Gujarati, Tamil, English)              |
+-----------------------------------------+-----------------------------------------+
                                          |
                                          v
+-----------------------------------------------------------------------------------+
|                        CORE FINANCIAL REASONING ENGINE                            |
|                                                                                   |
|  [ 1. Financial Pulse ]       [ 2. Safe-to-Spend™ ]      [ 3. Virtual Earmark ]   |
|   0-100 Holistic Health        Liquid Balance minus        Zero-risk planning     |
|   Velocity & Debt Metric       Future Commitments          without lock-in fees   |
+-----------------------------------------+-----------------------------------------+
                                          |
                                          v
+-----------------------------------------------------------------------------------+
|                     RESPONSIBLE NEXT BEST ACTION (NBA) GUARD                      |
|                                                                                   |
|   OFFER (Buffer)  |  ASSIST (Restructure)  |  WARN (Pause)  |  DO_NOTHING (Safety)|
|   >>> Commercial credit is PROGRAMMATICALLY SUPPRESSED during financial distress <<< |
+-----------------------------------------+-----------------------------------------+
                                          |
                                          v
+-----------------------------------------------------------------------------------+
|                       REGULATORY & COMPREHENSION SHIELD                           |
|                                                                                   |
|   * Standardized RBI Key Facts Statement (KFS)                                    |
|   * Audio Vernacular Breakdown of Total Cost of Credit                            |
|   * Interactive Borrower Comprehension Quiz before OTP Authorisation              |
|   * Statutory 3-Day Penalty-Free Cool-off Cancellation Window                     |
+-----------------------------------------------------------------------------------+

Prevents accidental overspending and bounced commitments without locking user liquidity in rigid bank deposits. $$\text{Safe-to-Spend} = \text{Liquid Balance} + \text{Confirmed Inflows} - \sum_{\text{confirmed}} (\text{Commitment Amount} - \text{Saved Amount}) - \text{Safety Buffer}$$

• Virtual Earmarking: Allocate funds for "Daughter's School Fees" or "Shop Inventory" without penalties or freezing accounts.
• Dynamic Cashflow Adjustment: Instantaneous recalculation upon any incoming UPI transfer or outgoing expense.

• Natural Voice Input: Speak freely in native dialects (e.g., "मेरी बेटी की कॉलेज फीस अगले महीने पचास हजार है").
• Multilingual Intent Parsing: Automatically categorizes commitments, extracts monetary values, maps target dates, and normalizes colloquial numbers (e.g., "पचास हज़ार" $\rightarrow$ ₹50,000).
• Explicit Verbal Confirmation: Never executes an earmark or transaction without repeating the terms in the user's native tongue and confirming consent.

A stateful ethical decision engine that puts customer safety above fintech commissions:

• OFFER: Recommends liquid emergency buffers and micro-savings only when cashflow is healthy.
• ASSIST: Guides users to restructure payment dates when commitments overlap.
• WARN: Flags impending overdrafts and recommends pausing non-essential expenses.
• PROTECT: Instantly alerts against suspicious merchant anomalies and freezes credentials.
• DO_NOTHING: The Golden Rule — Strictly halts all loan marketing and commercial credit offers when the customer's Financial Pulse indicates distress.

• Standardized KFS Breakdown: Plain-text breakdown of Principal, Processing Fees, Net Disbursal, APR, and Total Repayment Amount.
• Voice Explainability: Plays an Indic audio narration explaining the exact cost of borrowing.
• Mandatory Comprehension Check: An interactive 2-question quiz ensuring the borrower understands their monthly EMI commitment before disbursal.
• 3-Day Statutory Cool-off: Allows users to cancel the loan with zero penalty within 72 hours, adhering to RBI digital lending guidelines.

• Vector Retrieval: Embeds RBI digital lending master directions, government agricultural subsidies, and PM micro-credit schemes into Pinecone.
• Contextual Grounding: Provides cited, hallucination-free advice in response to user queries about rights, interest rate ceilings, and financial safety.

• Framework: Next.js 16.3.3 (App Router, Turbopack, React Server Components)
• Frontend Core: React 19, TypeScript 5.7
• Styling & Design System: Tailwind CSS v4, @tailwindcss/postcss, tw-animate-css
• UI Primitives: Base UI, Shadcn UI, Lucide React
• Artificial Intelligence:
  • Reasoning: Google Gemini 2.0 Flash (@google/genai)
  • Indic Voice & TTS: Sarvam AI API (Saaras Indic Speech-to-Text & Bulbul TTS)
  • Vector Retrieval: Pinecone Database (@pinecone-database/pinecone)
• Testing & Quality Assurance:
  • Native Node.js Test Runner (node --experimental-strip-types)
  • End-to-End Persona Verification Suites
  • Playwright Test Integration (@playwright/test)

• Node.js: v20.x or v22.x recommended
• Package Manager: npm, pnpm, or yarn

git clone https://github.com/dare-devil-coder/Vanifi.git
cd Vanifi

npm install
# or
pnpm install

Copy the .env.example file to .env.local:

cp .env.example .env.local

Fill in your API credentials:

GEMINI_API_KEY=your_gemini_api_key_here
SARVAM_API_KEY=your_sarvam_api_key_here
PINECONE_API_KEY=your_pinecone_api_key_here

npm run dev

Open http://localhost:3000 in your browser.

Vani-Fi includes automated regression and mathematical verification test suites:

Runs 15 automated test assertions covering Safe-to-Spend algebra, Pulse scoring, NBA suppression, and Indic voice parsing:

npm test

Simulates the complete user scenario from vernacular voice speech input to safe-to-spend recalculation, stressed state credit suppression, and RBI KFS loan authorization:

npm run test:e2e

npm run typecheck
npm run build
npm run start

• Anti-Predatory AI: First fintech engine that treats suppressing loans as a primary AI objective function.
• Vernacular Dialect Normalization: Translates colloquial Indic numeric constructs ("डेढ़ लाख", "पचास हज़ार") directly into validated financial models.
• Total Regulatory Alignment: Complies natively with Reserve Bank of India (RBI) Digital Lending Directives, 2022 and 2024 Fair Lending Practices.

Designed and developed with ❤️ for Bharat by dare-devil-coder for Hackout 2026. All rights reserved under the MIT License.