AI Accounting Pipeline

A domain-driven Django backend for an AI accounting automation platform - document ingestion, a two-pass LLM pipeline, and reconciliation-ready publishing to Xero and QuickBooks.

Two-pass LLM document pipeline (extraction + enrichment) with structured outputs and schema validation - more accurate VAT classification and category mapping.
Multi-provider LLM orchestration on LiteLLM - OpenAI, Vertex AI (Gemini) and Anthropic - with fallback routing, retries and rate-limit handling.
Async pipelines on Celery, Redis, AWS SQS and Lambda, with sub-document parallelisation.
pgvector embeddings for supplier intelligence and similarity search.
LLM observability: structured logging, step-level tracing and error classification.
Timeouts, retries, circuit breakers and rate limiting; secrets moved to AWS Secrets Manager.

Company: EazyCapture
Role: Senior Software Engineer / DevOps
Location: London, England
Timeline: 2025 – Now
Discipline: AI Products

Built with

Python
Django
LiteLLM
Celery
AWS
pgvector
DDD