AI Extraction Engine
The core magic of Outlay is our multi-modal AI extraction engine. It takes raw, unstructured images or PDFs and turns them into perfectly formatted accounting data.
How it Works
Traditional OCR (Optical Character Recognition) often fails on Indian receipts because of faded thermal ink, chaotic layouts, and bilingual text. Outlay bypasses standard OCR entirely.
Ingestion & Pre-processing
When a receipt is uploaded (via Web, WhatsApp, or Email), the image is enhanced for contrast. If it is a PDF, it is rasterized securely in memory.
Multi-Modal Vision Analysis
The image is passed to a fine-tuned Vision Language Model (LLM) that acts as a structural parser. It physically "reads" the layout of the receipt to understand context (e.g., distinguishing the subtotal from the grand total).
Structured JSON Output
The model guarantees a strict JSON schema containing the Merchant Name, Date, Currency, Total, Taxes, and the GSTIN.
Supported Document Types
| Format | Common Source | Accuracy |
|---|---|---|
| JPEG / PNG | Restaurant Bills, Taxi Receipts (Phone Camera) | 98% |
| Digital PDF | MakeMyTrip, AWS, Software Subscriptions | 99.9% |
| Scanned PDF | Bulk accountant uploads | 96% |
| HTML Body | Uber / Swiggy email receipts | 99.9% |
Handwritten Receipts
While the engine can read clear handwriting, heavily scribbled `kachcha` (raw) bills from unregistered merchants may require manual verification.