Create README.md

README.md

@@ -0,0 +1,445 @@
+---
+language:
+- en
+- hi
+license: apache-2.0
+size_categories:
+- 10K<n<100K
+task_categories:
+- text-classification
+- table-question-answering
+- text-generation
+tags:
+- finance
+- synthetic
+- banking
+- india
+- transactions
+- bank-statements
+- document-ai
+pretty_name: Indian Bank Statement Synthetic Dataset
+---
+
+# Dataset Card for Indian Bank Statement Synthetic Dataset
+
+This dataset contains synthetically generated Indian bank statements with realistic transaction patterns, merchant names, regional variations, and proper banking workflows representative of the Indian financial ecosystem. Available in both **scanned PDF** and **digital (structured JSON)** formats.
+
+## Dataset Details
+
+### Dataset Description
+
+This is a comprehensive synthetic dataset of Indian bank transactions designed to reflect realistic banking behaviors across multiple Indian banks, payment systems (UPI, NEFT, IMPS, RTGS), and transaction types. The dataset incorporates regional naming patterns, realistic transaction flows, running balance calculations, and India-specific banking features such as UPI reference numbers, IFSC codes, MICR codes, and merchant identifiers commonly seen in Indian bank statements.
+
+The dataset includes both **Current Accounts** (business banking) and **Savings Accounts** (individual banking) with transactions in two statement formats:
+- **Separate Debit/Credit Columns**: Traditional format with distinct debit and credit columns
+- **Single Transaction Column**: Combined format where debits and credits appear in one column with +/- indicators
+
+Each statement is provided in:
+- **Scanned PDF format**: Visual representation mimicking actual bank statement PDFs (suitable for OCR and document understanding tasks)
+- **Digital JSON format**: Structured data with rich metadata including account details, branch information, and transaction records
+
+**Note:** This dataset contains only legitimate transactions. It does NOT include fraudulent transactions or fraud patterns.
+
+- **Curated by:** AgamiAI Inc.
+- **Funded by:** AgamiAI Inc.
+- **Language(s):** English (primary), Hindi (romanized merchant/location names)
+- **License:** Apache 2.0
+
+### Dataset Sources
+
+- **Repository:** https://huggingface.co/datasets/agami-ai/indian-bank-statements
+- **Company Website:** https://www.agami.ai
+
+## Uses
+
+### Direct Use
+
+This dataset is suitable for:
+- **Document AI and OCR training**: Extract text and tables from scanned bank statement PDFs
+- **Information Extraction**: Train models to identify and extract key fields (account numbers, balances, transaction details)
+- **Transaction categorization and classification**: Classify transactions by type, merchant category, or purpose
+- **Financial document understanding**: Build systems that comprehend bank statement structure and semantics
+- **Chatbot and copilot training**: Train financial assistants to answer questions about bank statements
+- **Data processing pipeline testing**: Validate ETL systems for banking data
+- **Table extraction and parsing**: Train models to extract tabular transaction data from PDFs
+- **Named Entity Recognition (NER)**: Identify merchant names, locations, and banking entities
+- **Educational purposes**: Fintech and data science coursework
+- **Agentic AI development**: Train private AI agents for financial document processing workflows
+
+### Out-of-Scope Use
+
+This dataset should NOT be used for:
+- **Fraud detection or anti-money laundering (AML)**: Dataset does not contain fraudulent patterns
+- **Production compliance or regulatory reporting**: This is not real financial data
+- **Training models for actual credit decisions**: Lacks real creditworthiness signals
+- **Assuming complete representation** of all Indian demographics, regions, or banking behaviors
+- **Real-world anomaly detection**: Synthetic anomalies may not match real-world patterns
+
+## Dataset Structure
+
+### Statement Formats
+
+The dataset includes two transaction column formats:
+
+**Format 1: Separate Debit/Credit Columns (Traditional)**
+| Date | Description | Debit | Credit | Balance |
+|------|-------------|-------|--------|---------|
+| 01/01/2024 | UPI-Swiggy | 450.00 | - | 25,780.50 |
+| 02/01/2024 | NEFT Salary Credit | - | 50,000.00 | 75,780.50 |
+
+**Format 2: Single Transaction Column (Combined)**
+| Date | Description | Transaction | Balance |
+|------|-------------|-------------|---------|
+| 01/01/2024 | UPI-Swiggy | -450.00 | 25,780.50 |
+| 02/01/2024 | NEFT Salary Credit | +50,000.00 | 75,780.50 |
+
+### JSON Data Structure
+
+Each statement includes a comprehensive JSON file with the following structure:
+
+```json
+{
+  "bank_name": "Paramount Banking Corporation",
+  "account_holder": "CYIENT TECHNOLOGIES",
+  "account_holder_address": "F-346\nThird Floor\nHinjewadi\nPune\nMaharashtra\n520018",
+  "account_number": "90823789756",
+  "ifsc_code": "PARA0761987",
+  "micr_code": "899946557",
+  "branch_name": "PUNE HINJEWADI",
+  "branch_code": "6738",
+  "branch_phone": "8647919953",
+  "account_type": "CURRENT ACCOUNT- GENERAL",
+  "currency": "INR",
+  "customer_id": "134743833",
+  "opening_balance": 158458.03,
+  "closing_balance": 64424.49,
+  "start_date": "2024-01-01",
+  "end_date": "2024-03-31",
+  "statement_date": "2025-11-20",
+  "interest_rate": 2.83,
+  "transactions": [...]
+}
+```
+
+### Transaction Record Structure
+
+Each transaction in the `transactions` array contains:
+
+```json
+{
+  "date": "2024-01-01 12:40:40",
+  "value_date": "2024-01-01",
+  "description": "NEFT Dr-471179370408-HDFC0009038-RIDDHI RAVAL",
+  "cheque_no": "862512",
+  "debit": 13932.79,
+  "credit": null,
+  "balance": 144525.24,
+  "branch_code": "3421",
+  "failed": false
+}
+```
+
+### Data Fields
+
+**Statement-Level Metadata:**
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `bank_name` | string | Name of the bank issuing the statement |
+| `account_holder` | string | Name of account holder (individual or business) |
+| `account_holder_address` | string | Complete address with line breaks |
+| `account_number` | string | Bank account number |
+| `ifsc_code` | string | Indian Financial System Code (11 characters) |
+| `micr_code` | string | Magnetic Ink Character Recognition code |
+| `branch_name` | string | Name and location of branch |
+| `branch_code` | string | Branch identifier code |
+| `branch_phone` | string | Branch contact phone number |
+| `account_type` | string | Account type (Savings/Current, with sub-type) |
+| `currency` | string | Currency (INR for all records) |
+| `customer_id` | string | Bank's internal customer identifier |
+| `opening_balance` | float | Account balance at statement start |
+| `closing_balance` | float | Account balance at statement end |
+| `start_date` | string | Statement period start date (YYYY-MM-DD) |
+| `end_date` | string | Statement period end date (YYYY-MM-DD) |
+| `statement_date` | string | Date statement was generated |
+| `interest_rate` | float | Current interest rate (% per annum) |
+
+**Transaction-Level Fields:**
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `date` | string | Transaction date and time (YYYY-MM-DD HH:MM:SS) |
+| `value_date` | string | Value date (when funds cleared) |
+| `description` | string | Full transaction description with bank codes and merchant info |
+| `cheque_no` | string | Cheque number (empty string if not applicable) |
+| `debit` | float | Debit amount in INR (null if credit transaction) |
+| `credit` | float | Credit amount in INR (null if debit transaction) |
+| `balance` | float | Running account balance after transaction |
+| `branch_code` | string | Branch code where transaction occurred |
+| `failed` | boolean | Transaction failure status (false for successful, true for failed/reversed) |
+
+### Transaction Types Included
+
+- **UPI (Unified Payments Interface)**: UPI/DR, UPI/CR with reference numbers
+- **NEFT (National Electronic Funds Transfer)**: NEFT Dr, NEFT Cr with bank codes
+- **RTGS (Real Time Gross Settlement)**: RTGS Dr, RTGS Cr for high-value transfers
+- **IMPS (Immediate Payment Service)**: IMPS Dr, IMPS Cr, IMPS Salary Transfers
+- **Cheque Transactions**: Chq Paid, By Clg (Clearing)
+- **Cash Transactions**: Cash Withdrawal, Cash Deposit (CASH-BNA-SELF)
+- **ATM Transactions**: ATM WDL (Withdrawal)
+- **Service Charges**: Various bank fees (online banking, statement charges, forex markup)
+- **Reversals**: Failed transaction reversals with REVERSAL prefix
+
+### Account Types
+
+- **Savings Accounts**: Individual banking with lower transaction volumes
+- **Current Accounts**: Business banking with higher transaction volumes and no transaction limits
+
+### Data Splits
+
+The dataset is organized into train, validation, and test splits to support machine learning workflows. Specific split sizes are available in the dataset repository.
+
+## Dataset Creation
+
+### Curation Rationale
+
+India has one of the world's fastest-growing digital payment ecosystems, with UPI processing billions of transactions monthly. However, publicly available datasets for training AI models on Indian financial documents are scarce due to privacy and regulatory constraints. 
+
+AgamiAI created this synthetic dataset to support the development of privacy-preserving, accurate AI solutions for financial services. As a company specializing in private AI agents for enterprise clients, particularly in financial services, AgamiAI recognized the critical need for high-quality training data that:
+
+1. Enables development and testing of document AI systems for Indian bank statements
+2. Supports OCR and information extraction model training on scanned financial documents
+3. Provides realistic training data reflecting India-specific payment systems and banking formats
+4. Allows developers to build and test banking applications without accessing real customer data
+5. Includes both scanned (unstructured) and digital (structured) formats for comprehensive document understanding tasks
+6. Supports research in transaction classification, document parsing, and financial NLP
+7. Facilitates the development of agentic AI workflows for financial document processing
+
+### Source Data
+
+#### Data Collection and Processing
+
+This is a **fully synthetic dataset** with no real customer information. The generation process leverages AgamiAI's expertise in building enterprise-grade AI solutions and includes:
+
+**Statement Generation:**
+- Two format types: separate debit/credit columns and single combined transaction column
+- Both scanned PDF (for OCR tasks) and structured JSON (for direct data processing)
+- Realistic bank statement templates matching actual Indian bank formats
+- Proper letterheads, logos, and formatting (synthetic bank brands)
+
+**Transaction Generation:**
+- Probabilistic modeling of realistic transaction patterns (frequency, amounts, timing)
+- Proper debit/credit flows with accurate running balance calculations
+- Transaction type distribution matching Indian banking patterns (high UPI usage, business-focused NEFT/RTGS)
+- Salary credits, vendor payments, cash management at realistic intervals
+- Transaction reversals and failed transactions for realistic edge cases
+
+**Indian Banking Features:**
+- UPI reference numbers following standard 12-digit formats
+- NEFT/RTGS reference numbers with bank codes (HDFC, ICICI, Citi, etc.)
+- Realistic business and individual names across Indian regions
+- IFSC codes following standard format (BANK0123456)
+- MICR codes (9 digits)
+- Branch codes and locations
+- Service charges and bank fees
+
+**Account Variations:**
+- Current Accounts: Business entities (companies, partnerships)
+- Savings Accounts: Individual account holders
+- Various transaction volumes (low to high frequency)
+- Different balance ranges (small to large accounts)
+
+**Regional Coverage:**
+- Major metros: Mumbai, Delhi, Bangalore, Pune, Chennai, Kolkata, Hyderabad
+- Business entities: IT companies, manufacturing firms, retail chains, financial services
+- Mix of B2B transactions (business-to-business) and individual transactions
+
+**Temporal Patterns:**
+- Quarterly statement periods (3-month spans)
+- Monthly salary/revenue patterns for businesses
+- Vendor payment cycles
+- Service charge applications (monthly/quarterly)
+- Weekend vs weekday transaction patterns
+
+#### Who are the source data producers?
+
+This is entirely synthetic data generated algorithmically by AgamiAI Inc. No real individuals, businesses, banks, or merchants contributed actual transaction data.
+
+### Annotations
+
+Transaction types and metadata were assigned algorithmically based on transaction patterns:
+- **Transaction Type Classification**: UPI, NEFT, RTGS, IMPS, Cheque, ATM, Cash automatically tagged
+- **Entity Extraction**: Merchant names, bank names, reference numbers systematically generated
+- **Temporal Features**: Date, value_date, and statement periods logically consistent
+
+#### Personal and Sensitive Information
+
+**This dataset contains NO real personal or financial information.** All elements are synthetically generated:
+- Account numbers: Fictional/masked
+- Business names: Generated (mix of real company name patterns and fictional entities)
+- Individual names: Generated using Indian naming patterns
+- Phone numbers: Synthetic (10-digit format)
+- Addresses: Fictional but realistic (actual area/city names with fictional building/street)
+- IFSC codes: Synthetic (following standard format)
+- MICR codes: Fictional
+- Transaction amounts: Statistically modeled
+- Balances: Generated based on transaction flows
+- Branch details: Fictional branches with realistic naming
+
+No real individuals or businesses can be identified from this data.
+
+## Bias, Risks, and Limitations
+
+**Known Limitations:**
+
+1. **No Fraud Patterns**: Dataset contains only legitimate transactions - NOT suitable for fraud detection training
+2. **Urban/Business Bias**: Reflects urban business banking behaviors more than rural or very small-scale individual banking
+3. **Transaction Volume**: Business current accounts may show different patterns than retail savings accounts
+4. **Regional Coverage**: While multi-regional, may not capture all linguistic and business variations across India's states
+5. **Temporal Simplification**: Seasonal business patterns simplified compared to real-world complexity
+6. **Document Variations**: Scanned PDFs may not capture all possible bank statement layouts and formats used across Indian banks
+7. **OCR Challenges**: Scanned documents generated synthetically may not include all real-world OCR challenges (handwriting, stamps, poor scans)
+
+**Technical Limitations:**
+
+- Transaction description formats standardized; real statements have more variation
+- Failed/reversed transactions simplified compared to real-world complexity
+- Cross-border transactions limited or excluded
+- Does not include all possible service charges and bank fees
+- Statement formats limited to common layouts (not exhaustive of all Indian banks)
+
+**Social and Ethical Considerations:**
+
+- Dataset reflects formal banking sector; excludes informal financial systems
+- Business transactions may not represent individual consumer spending patterns
+- Modern digital payment heavy; traditional banking methods (cash, cheques) represented but lower frequency
+- Should not be used to make assumptions about real businesses' or individuals' financial behaviors
+
+### Recommendations
+
+**For Model Developers:**
+- Use for document AI, OCR, and information extraction training
+- Validate extraction models on real anonymized data before production
+- This is suitable for structure and format learning, not for behavioral modeling
+- **Do NOT use for fraud detection** - lacks fraudulent transaction patterns
+- Consider using AgamiAI's platform for deploying privacy-preserving AI models trained on this data
+
+**For Researchers:**
+- Clearly disclose use of synthetic data in publications
+- Focus research on document understanding, not financial behavior
+- Validate findings with real data where possible
+- Consider this for algorithm development, not financial insights
+
+**For Banking/Fintech Applications:**
+- Excellent for testing document processing pipelines
+- Use for UI/UX testing with realistic-looking statements
+- Good for training staff on document review workflows
+- Do NOT use for actual financial analysis or compliance
+- Validate regulatory requirements with real anonymized data
+- For production deployment of AI solutions, consider AgamiAI's private AI platform for secure, compliant deployment
+
+**For Document AI Tasks:**
+- Train table extraction models on the scanned PDF format
+- Use JSON for ground truth validation
+- Test entity recognition and classification systems
+- Benchmark OCR accuracy across different statement formats
+
+## Citation
+
+**BibTeX:**
+
+```bibtex
+@dataset{indian_bank_statement_synthetic_2025,
+  author = {AgamiAI Inc.},
+  title = {Indian Bank Statement Synthetic Dataset},
+  year = {2025},
+  publisher = {HuggingFace},
+  url = {https://huggingface.co/datasets/agami-ai/indian-bank-statements}
+}
+```
+
+**APA:**
+
+AgamiAI Inc. (2025). *Indian Bank Statement Synthetic Dataset* [Data set]. HuggingFace. https://huggingface.co/datasets/agami-ai/indian-bank-statements
+
+## Glossary
+
+**Indian Banking Terms:**
+
+- **UPI (Unified Payments Interface)**: India's instant real-time payment system, the most popular digital payment method
+- **NEFT (National Electronic Funds Transfer)**: Batch processing system for interbank transfers (half-hourly settlements)
+- **RTGS (Real Time Gross Settlement)**: Real-time bank transfer system for high-value transactions (typically ₹2 lakh+)
+- **IMPS (Immediate Payment Service)**: Instant interbank transfer service, 24/7 availability
+- **IFSC Code**: Indian Financial System Code - unique 11-character code identifying bank branches (e.g., HDFC0001234)
+- **MICR Code**: Magnetic Ink Character Recognition code - 9-digit code for cheque processing
+- **Current Account**: Business/commercial account with no transaction limits, no interest
+- **Savings Account**: Individual account with transaction limits, earns interest
+- **Value Date**: Date when funds are actually debited/credited (may differ from transaction date)
+- **Reversal**: Failed transaction that was initially processed but later reversed
+
+**Document Formats:**
+
+- **Scanned PDF**: Image-based PDF mimicking scanned bank statements (for OCR training)
+- **Digital JSON**: Structured data format with all statement and transaction details
+- **Separate Columns Format**: Traditional format with distinct Debit and Credit columns
+- **Single Column Format**: Combined format where transactions show +/- in one column
+
+## More Information
+
+### About AgamiAI
+
+AgamiAI Inc. builds private AI solutions for enterprise clients, with a focus on industries where privacy, accuracy, and compliance are non-negotiable. Our platform delivers:
+
+- **Private by Design**: AI models fine-tuned with your data, deployed securely in your cloud
+- **Agentic AI**: Adaptive agents for documents, research, insights, and workflow automation
+- **Enterprise-Grade**: Built for accuracy, compliance, and scalability with secure deployment
+- **Industry Focus**: Specialized solutions for Finance, Healthcare, Legal, Consulting, and Research
+
+AgamiAI's team brings deep experience from companies like Google, Meta, and Airtable, with a mission to help enterprises turn AI into real business impact while maintaining trust, precision, and control over their data.
+
+Visit us at: **https://www.agami.ai**
+
+### File Structure
+
+Each statement in the dataset includes:
+- `[statement_id].pdf` - Scanned bank statement (PDF format)
+- `[statement_id].json` - Structured data (JSON format with full metadata)
+
+### Validation Approach
+
+Quality was validated through:
+- JSON schema validation for all structured data
+- Balance calculation verification (running balances mathematically correct)
+- Format consistency checks across scanned and digital versions
+- Expert review by professionals
+- Cross-validation between PDF and JSON content
+
+### Future Updates
+
+Planned enhancements may include:
+- Additional regional merchant diversity
+- More bank formats and statement styles
+- International transaction patterns
+- Investment and trading transactions
+- Loan and credit card statement formats
+- Fraudulent transactions
+
+## Dataset Card Authors
+
+AgamiAI Inc.
+
+## Dataset Card Contact
+
+For questions, feedback, or collaboration opportunities:
+- **Website**: https://www.agami.ai
+- **Email**: Contact us through our website
+- **HuggingFace**: https://huggingface.co/agami-ai
+
+---
+
+**Version:** 1.0.0  
+**Last Updated:** November 2025
+
+**License:** Apache 2.0
+
+**Privacy Notice:** This dataset contains entirely synthetic data. No real personal or financial information is included.