Data Processing Practices

1. Purpose and Scope

This Data Processing specification describes the technical, operational, and architectural mechanisms through which ChatGraphs collects, parses, transforms, analyzes, and stores financial records submitted by users. Our architecture is designed with tenant isolation, deterministic financial mathematics, and user data ownership as foundational constraints.

2. Ingestion & File Parsing

When a user uploads a financial statement, trial balance, or spreadsheet (.xlsx, .xls, .csv, or .json):

• Temporary Memory Staging: The file payload is streamed over an encrypted TLS connection directly to our secure processing pipeline.
• Tabular Parsing: Structural parsers extract sheet names, rows, columns, date intervals, and numerical values without executing embedded macros or untrusted active code.
• Data Sanitization: Input fields are checked for arithmetic consistency and validated against standard debit-credit balancing requirements.

3. Account Classification & Structural Mapping

ChatGraphs applies semantic and rule-based classification to organize raw ledger entries into standardized financial presentation hierarchies (such as Companies Act Schedule III groupings, non-corporate schedules, or custom chart-of-accounts):

• Classification applies non-destructive tags to accounts (e.g. classifying “SBI Current Account” as “Cash & Bank”).
• The original ledger codes and historical values remain intact and verifiable.
• Users maintain the ability to review, customize, or override any system-suggested mapping at any time.

4. Mathematical Computations & Ratio Intelligence

All mathematical calculations—including Compound Annual Growth Rates (CAGR), DuPont Return on Equity breakdowns, variance bridges, and liquidity ratios—are executed through deterministic financial formulas in transient memory. Outputs are mathematically auditable against the underlying source ledgers.

5. AI Inference & Context Isolation

When a natural language question is submitted (e.g., “Why did EBITDA margin decline in FY24?”):

• Only the relevant summarized financial metrics and question text are passed to the inference context.
• Queries are processed ephemerally to generate explanations and chart configurations.
• Your proprietary financial statements and transaction records are never used to train public foundation AI models.

6. Encryption & Multi-Tenant Isolation

ChatGraphs maintains strict isolation boundaries between customer accounts:

• In Transit: All HTTP traffic and API communications are encrypted using TLS 1.2 or TLS 1.3 with modern cipher suites.
• At Rest: Financial data stored in databases or file caches is encrypted using AES-256 standards.
• Tenant Boundaries: Data queries are logically scoped to authenticated workspace identifiers, preventing cross-tenant access.

7. Data Retention, Portability & Erasure

• User Portability: Users can export their financial data, structured statements, and generated visualizations at any time in standard formats (PDF, Excel, PNG, CSV).
• Instant Erasure: When a user deletes a file, conversation, or workspace, the data is immediately removed from active serving layers and permanently purged from backup snapshots within 30 days.

8. Inquiries & Data Subject Requests

If you have technical questions regarding our data processing pipelines or wish to request data verification, please submit an inquiry via our Contact Page or review our Security Practices.