Vector candidate recall
GPU similarity search surfaces plausible matches across millions of lines.
Reconn resolves bank-to-ledger reconciliation as a GPU vector + graph problem — fuzzy-matching millions of lines across timing, partials, and FX, then proposing the journal entry.
Part of the Closora close
Reconn resolves bank-to-ledger reconciliation as a GPU vector + graph problem — fuzzy-matching millions of lines across timing, partials, and FX, then proposing the journal entry.
Every product in the suite shares memory, audit trail, and the verifier — composing one autonomous close function from owned models.
GPU similarity search surfaces plausible matches across millions of lines.
Many-to-many matching via accelerated graph algorithms (cuGraph).
A fine-tuned model scores hard matches and drafts the entry.
Unmatched items route to the exception agent with reasons.
Reconn receives its inputs from your ERP, banks, or document stores in real time.
Owned, fine-tuned models make a structured, tool-callable decision with rationale.
An independent verifier gates the action against policy before anything is written.
The outcome and its provenance land in the immutable audit timeline.
| Compute | RAPIDS + cuGraph + cuVS |
|---|---|
| Hardware | DGX A100 / H100 |
| Speed | Days → minutes |
| Serving | TensorRT match scorer |
Reconn handles multi-bank bank reconciliation end to end, escalating only true exceptions.
Reconn handles cash application end to end, escalating only true exceptions.
Reconn handles credit-card / expense recon end to end, escalating only true exceptions.
Reconn included with the Growth platform.
Private fine-tune of Reconn on your corpus.
The fine-tuned GAAP-reasoning and independent verifier engine behind every decision.
View product →GPU-served, layout-aware extraction for invoices, POs, and remittances.
View product →We'll run it in shadow mode against a recent period so you can compare.