Ledgerline

SME finance teams spend their month chasing statements, typing invoice numbers and reconciling M-Pesa flows against bank flows by hand. The work is exacting, joyless and error-prone — and it delays the one report leadership actually reads.

An ingestion-first operations console: bank statements, M-Pesa CSVs, invoices and receipts flow in (email, upload, folder watch) and are extracted into structured entries by a document pipeline. A matching engine reconciles entries against the ledger; confident matches settle silently, uncertain ones land in a human-first exception queue with the evidence attached.

• ▸Queue-based ingestion pipeline (BullMQ) — every document versioned
• ▸OCR + LLM extraction into a typed entries schema with confidence scores
• ▸Rules engine for deterministic matches; embeddings for fuzzy dedupe
• ▸PostgreSQL ledger with immutable audit trail; object storage for source documents
• ▸Approval flows with maker-checker separation for exceptions

A worklist, not a dashboard: the finance officer opens Ledgerline to a prioritised queue of exceptions, each with the source document, the extracted entry and the reason it paused. Clear the queue; close the month.

Extraction confidence scores route work — high confidence auto-posts, the rest goes to humans. Cash-flow forecasting from the reconciled ledger, with the assumptions on display rather than hidden.

Targets are what this design is dimensioned for — not results claimed from work we haven't done. We think you should hold us to exactly this standard of honesty.

• Design target — 80% of entries auto-matched without human touch
• Design target — every posting traceable to its source document
• Design target — month-end close in days, not weeks