How BasilTax works
BasilTax connects messy documents to a deterministic tax engine. LLMs help read and structure; math is not “guessed” by a model.
1. Ingestion
PDFs are parsed with text extraction; images may use OCR in our Docker environment. CSV uploads bypass vision steps.
2. PII-safe extraction
Text is redacted, then passed to structured extraction prompts with Pydantic schemas. Fields map to internal keys aligned with Form 1040 line mapping.
3. Ledger normalization
Extracted amounts flow into a chart-of-accounts style ledger so debits and credits reconcile before tax forms are filled.
4. Deterministic computation
Federal and supported state engines compute brackets, credits, Schedule D, Schedule E, QBI, and more—using the same inputs you can inspect per document.
5. Chat with citations
Questions like “Why is line 16 this amount?” retrieve context from your documents and rules knowledge, with citations to filenames and pages where available.
6. Audit pack
Export a PDF that bundles traces suitable for preparer review—or create a redacted share link (see Security).