Jeffrey Preston

I build the part of a fintech product where a wrong number costs money: payment rails and custodial wallets across Bitcoin, Ethereum/ERC-20, Tron/TRC-20 and TON, double-entry ledgers that reconcile against the chain, and pipelines that turn bank statements and receipts into posted entries.

What I have shipped:

- A ledger core in Go that two unrelated products adopted unchanged, which is the real test of whether an accounting model is right.

- A payout rail where every batch dry-runs first: it shows the fee and the exact debit per row before anything is sent, and lists the rows that will not go out with the reason and the shortfall, by line number in the file.

- A reserve reconciliation that compares obligations to users against on-chain balances every five minutes and freezes withdrawals for an asset automatically when the solvency invariant breaks.

- A financial-model backend: a two-year monthly plan by category where dragging one month's ceiling recalculates the rest. 313 endpoints, 9,333 tests, in daily production, maintained by one person.

Go, Python, TypeScript, PostgreSQL, Docker, Kubernetes. What I optimise for is not speed but provability: that the balances agree afterwards, and that a discrepancy can be explained.

Available for contract work, remote.