Moneio

Inspiration

We wanted an "AI CFO" you can talk to like a teammate—fast answers on cash flow, burn, and runway without clicking through dashboards. ElevenLabs gives it a natural voice; Gemini provides the financial reasoning.

What it does

Voice-first finance assistant: users speak questions, it listens, queries live financial data, reasons with Gemini, and replies in human-sounding ElevenLabs speech. It handles cash flow, burn rate, runway, spending breakdowns, and quick summaries—all hands-free.

How we built it

• Frontend: Next.js with an ElevenLabs-powered voice widget for ASR/TTS and a simple state machine (listening → thinking → speaking).
• Backend: API route that calls Gemini/Vertex for reasoning, hits our finance data, and returns concise replies for TTS.
• Data: Sample 2025 transactions plus live workspace data for consistent demos.
• Infra: Deployed on Google Cloud (Cloud Run) with environment-secured keys for Gemini and ElevenLabs.

Challenges we ran into

• Making voice UX snappy (low latency) while still running an LLM.
• Handling noisy transcripts and mapping them to the right financial queries.
• Ensuring responses stay concise enough for speech without losing accuracy.

Accomplishments that we're proud of

• End-to-end voice loop in under a few seconds from speech to answer to speech.
• Consistent financial answers grounded in our transaction data.
• A reusable prompt/response pattern that keeps answers short and TTS-friendly.

What we learned

• Prompt design for speech requires extra brevity and structure.
• ElevenLabs voices make LLM outputs feel far more usable than text alone.
• Latency budgeting across ASR → LLM → TTS is key for a good voice UX.

What's next for Moneio

• Multi-language voices and localized financial summaries.
• Push-to-talk hardware demo.
• Richer "why/how" explanations with charts and inline citations.
• Alerts and follow-up actions (e.g., "reduce burn by cutting X").

Built With

• elevenlabs-agents-&-tts
• google-gemini-(vertex-ai)
• next.js
• node.js
• prisma
• speech/asr
• supabase/postgres
• tailwind-css
• typescript
• vercel/cloud-run-(gcp)
• web