Voxfin is in development — join the design-partner programRequest access →
Finmaro voice interfaceDeveloping

Voxfin

Talk to your CFO. Get board-ready answers in seconds.

Voxfin is a realtime voice interface over the Finmaro financial model. Ask a question in the car, before a board call, or during a meeting. Voxfin grounds the answer in the live model, routes the numbers through Verifin, and preserves the transcript.

Grounded voice answersVerification before speechPersistent transcript
Voice briefing · runwayIllustrative session
What happens to runway if we hire five engineers in September?
VoxfinPhase the hires 2+3.

Hiring all five in September reduces median runway from 14.2 to 11.6 months. Phasing two in September and three after the October revenue checkpoint preserves 12.8 months.

✓ Verified against 7 sourcesTranscript saved

Illustrative voice flow. Realtime Riva pipeline is planned.

The access problem

Financial judgment should not require a dashboard session.

Founders make decisions while moving between meetings. Voxfin is designed for the moment when typing a query breaks the flow but the answer still needs to be grounded, precise, and auditable.

Conversation

Ask the question the way you think it.

Natural spoken questions can include follow-ups, interruptions, and context from the live conversation.

Grounding

Answer from the model, not memory.

Voxfin queries connected records and Finsim scenarios before forming a recommendation.

Audit

Keep the spoken decision reviewable.

Transcripts, sources, assumptions, and approvals remain available after the meeting ends.

How Voxfin works

A voice pipeline with a finance trust gate.

Listen

Stream the question

Capture speech with interruption handling and preserve the conversational context.

Ground

Call finance tools

Query the live model, connected records, and Finsim scenario engine.

Verify

Check the answer

Route consequential figures through Verifin before speech synthesis.

Speak

Deliver and record

Return a concise answer and preserve the transcript and source trail.

Planned technology

Low latency is part of the product.

A useful voice CFO needs streaming recognition, reasoning, verification, and speech synthesis to feel like one continuous conversation.

Development disclosure: Finmaro’s agent reasoning is the foundation. Streaming ASR/TTS, sub-300ms turns, multilingual speech, and edge deployment are planned.
SpeechRiva streaming ASR and TTSPlanned
ReasoningTensorRT-LLM low-latency inferencePlanned
ServingTriton multi-model pipelinePlanned
EdgeJetson Orin for regulated environmentsPlanned

Ask the financial question while it still matters.