LLM engineer: receipt and invoice extraction with human review
$75–$120/hr · AI & Machine Learning · posted by Ledgerline
Accounting firms upload thousands of receipts and supplier invoices to Ledgerline every day. Our template-based OCR breaks whenever a layout changes. We want an LLM-based extraction pipeline with confidence scores and a review queue.
You will:
• Build an evaluation set from 1,000 labelled documents we already have
• Design schema-constrained extraction with validation (totals, tax, dates)
• Route low-confidence fields to our existing review UI
• Measure accuracy, latency and cost per document
About you:
• You have shipped LLM features to production, not just prototypes
• Strong Python; experience with evaluation frameworks
• Pragmatic about cost and latency
Skills
Python, LLM Integration, Document AI, Evaluation, FastAPI