An AI analyzer that turns credit card PDFs into interactive dashboards.
Upload any statement — Amex, Chase, or regional banks. Gemini 2.0 Flash extracts every transaction with 99% logic precision, categorizes spending, renders interactive charts, and exports to CSV. No login, no data storage.
Credit card statements are PDF black boxes. Understanding your spending requires extracting data that was never designed to be extracted.
A person downloads their monthly credit card statement. It's a 12-page PDF: transactions in a formatted table, running totals at the bottom, category labels that differ by bank. To actually analyze their spending, they either read it manually or paste it into a spreadsheet — row by row, 80 transactions, every month.
The failures are structural. Bank PDFs aren't designed for data export — they're designed for printing. Extracting transactions with standard PDF tools produces garbled text when tables span pages, misaligned columns when formatting shifts, and silently dropped rows when the PDF is scanned rather than digital. By the time the data is clean, the insight is stale.
The root cause isn't user behavior. It's that PDF statement format was standardized for human reading, not machine analysis — and every bank has a different layout, different table structure, and different category taxonomy. A template-based extractor works for one bank and breaks on the next. When multi-bank support is required, templates don't scale.
StatementFlow uses Gemini 2.0 Flash to read statements the way a financial analyst would — understanding layout context, inferring transaction boundaries across page breaks, and classifying spending by merchant semantics, not column position. No templates. No format-specific rules to maintain.
PDF in. Transactions extracted. Dashboard rendered. CSV ready.
Gemini reads layout, not column positions. Multi-bank without templates.
pdfjs-dist renders each PDF page for Gemini 2.0 Flash to read as a visual and textual document. The model understands that a transaction boundary is a new row regardless of whether the bank uses lines, spacing, or indentation to separate them. Amex, Chase, regional bank formats — same extraction pipeline, no format-specific rules.
Spend categorization uses merchant semantics, not keyword lists.
"Amazon.com*1A2B3C4D" is categorized as Shopping, not Unknown, because the model understands merchant naming conventions. "SQ*BLUE BOTTLE SF" maps to Dining. Category assignments reflect how a financial analyst would read the statement, not how a pattern-matcher would keyword-scan it.
The dashboard renders while you review, not after you wait.
Recharts powers dynamic charts that update as transactions load. Category-wise spend breakdowns, monthly trend lines, and a fully searchable transaction table are generated from the extracted data without a page reload. The dashboard is the deliverable — not a data file the user has to chart themselves.
CSV export makes the data portable. No lock-in.
Every extracted transaction — date, merchant, amount, category — exports to a clean CSV on demand. Users can take it into Excel, their budgeting app, or their accountant's preferred tool. StatementFlow is the extraction and analysis layer, not a data silo.
Three things change once AI reads your statement.
Transaction extraction logic accuracy
Gemini 2.0 Flash reads statements the way a trained analyst would — understanding page-spanning tables, merged cells, and multi-line merchant names without a format-specific template.
From PDF upload to full spending dashboard
Extraction, categorization, and chart rendering complete in a single pass. A 12-page, 80-transaction statement produces a full dashboard faster than opening a spreadsheet.
Server-side document retention
Documents are processed in memory and sent to the Gemini API for extraction only. No server stores your statement. Extracted metadata lives in your browser's localStorage — private by architecture, not by policy.
Numbers observed in Brilworks' StatementFlow deployment at statementflow.brilworks.com. Extraction precision varies with PDF quality — scanned statements at low DPI perform below the 99% benchmark. Digital PDFs from major banks consistently hit or exceed it.
Honest fit criteria. We'd rather say no than oversell.
✓Strong fit if
- You need to analyze credit card spending across months or multiple cards without manual data entry
- Your statements come from multiple banks with different PDF formats
- You want spending insights in a shareable dashboard format, not a raw data dump
- Privacy is a hard requirement — processing in memory with no server storage is the right architecture for your use case
✗Not a fit if
- You need transaction data synced automatically from bank APIs rather than uploaded manually
- Your statements are low-resolution scans where OCR accuracy is already a problem before AI analysis
- You need multi-user access or cloud persistence for extracted data across devices
- You're building a compliance or tax product that requires legally auditable extraction provenance
Book a 30-minute scoping call.
We'll look at your statement formats, your analysis workflow, and your data privacy constraints — then tell you whether AI-powered extraction fits and what a production deployment for your specific bank coverage would require.
+ 91 9313644148
sales@brilworks.com
503, Fortune Business Hub, Science City Road, Sola, Ahmedabad, Gujarat, India Pincode- 380060
