TRACE: Voice-Native Business Investigation Agent

TRACE is a voice-native business investigation agent built for situations where something does not add up. Businesses rarely keep the evidence needed to explain a problem in one place. An order may exist in one system, an invoice in another, a POS transaction somewhere else, and the corresponding bank settlement may be missing, duplicated, delayed, or difficult to identify. Finding the real cause often means manually searching across disconnected records. TRACE lets a user simply describe the problem by voice: “Yesterday’s revenue is short by ₦210,000. Investigate it.” TRACE turns that request into an investigation. It identifies relevant records, traces relationships between transactions, builds an evidence trail, forms competing hypotheses, and tests those hypotheses against available data. Instead of presenting an unsupported AI answer, TRACE distinguishes between evidence, inference, and uncertainty. For example, TRACE may discover an order, invoice, delivery record, and POS transaction for ₦120,000 but no corresponding bank settlement. It can explain that the payment is likely unresolved while making clear that the available evidence does not prove why it is missing. The user can then ask, “Why do you think that?” or simply, “Prove it.” TRACE can also perform controlled actions such as flagging a discrepancy, creating a reconciliation task, adding an investigation note, or drafting a follow-up message. Consequential financial actions remain behind human approval. The voice interface is powered by AssemblyAI’s Voice Agent API, using its real-time conversational pipeline, turn detection, barge-in, and tool-calling capabilities. TRACE uses these capabilities as the interaction layer for an underlying investigation system rather than treating voice as a cosmetic feature. TRACE is not designed to replace accounting software. It is designed to answer a different question: When something doesn't add up, what actually happened, and can you prove it?