Back to Platform Overview
FinLens v2Text-to-SQL AuditorAnti-Injection Guard

Natural Language SQL Auditor

Converts complex natural language queries into secure, parameterized SQL. Upload bank statements in PDF or CSV formats to inspect transaction schemas, run custom SQL computations, and visualize financial distributions without injection risk.

3Statement formats normalized into one transaction schema
0Write operations allowed from generated SQL execution paths
15+Offline keyword routes for air-gapped financial analysis
100%Displayed queries use visible parameter bindings before execution

Natural language analysis with database discipline

FinLens turns messy account statements into auditable SQL workflows, keeping calculations grounded in the database instead of trusting free-form model arithmetic.

Statement Normalization

PDF and CSV inputs are mapped into canonical debit, credit, balance, channel, and timestamp fields before any question is answered.

Schema-Aware SQL Planning

The natural language layer works against visible table metadata, then generates parameterized SELECT-only queries for deterministic math.

Analyst-Ready Outputs

Answers include SQL, bound variables, tabular results, and chart-ready aggregates so reviewers can inspect both logic and result.

Bank Statement Source

Click to select or drag financial statement here

Supports CSV, pipe-delimited text, or bank PDFs

Database Dialect

Target Schema Browser

accounts4 rows
account_id (INT, PK)
account_number (VARCHAR)
balance (DECIMAL)
currency (VARCHAR)
transactions1420 rows
tx_id (INT, PK)
account_id (INT, FK)
amount (DECIMAL)
channel (VARCHAR)
timestamp (TIMESTAMP)
status (VARCHAR)

SQL Injection Protection: FinLens operates on dynamic tokenized mappings. Queries are structurally validated to prevent arbitrary statement executions. Safe, primary parameterized variables are generated via LLM mappings and strictly bound prior to DBMS execution.

From question to verified ledger answer

SELECT-only execution
01

Parse

Convert natural language into intent, filter clauses, grouping requirements, and requested metric type.

02

Plan

Match the request to known schema fields and block unsafe verbs before SQL creation.

03

Execute

Run a parameterized SELECT statement against PostgreSQL or SQLite fallback storage.

04

Explain

Return SQL, bindings, rows, and chart-friendly aggregates for reviewer confidence.