A financial analytics and portfolio management platform that turns scattered account data into decisions people can actually act on.
- React
- Node.js
- Python
- PostgreSQL
Verba
Every conversation, understood
A conversation intelligence platform that transcribes calls as they happen, clusters what customers are really asking about, and helps the person on the line while the call is still live.
- Industry
- Customer Experience, Business
- Year
- 2026
- Screens
- 09
Overview
About the project
Verba is a concept product for support and sales teams whose most valuable information is locked inside recorded calls nobody has time to listen to. It transcribes conversations in real time with speaker separation, labels intent and sentiment turn by turn, and turns each call into a structured summary with action items rather than a wall of text. The design problem was restraint: an assistant that interrupts an agent mid-call is worse than no assistant at all, so guidance is placed in a quiet side rail that suggests rather than demands, and every score or claim can be played back at the exact second it came from.
The challenge
Call recordings are the richest record of what customers want and the least used, because reviewing them costs more time than any team has. Manual quality checks sample a fraction of conversations, which means coaching is based on whichever calls happened to be picked. The harder design constraint was live assistance: anything that pulls an agent's attention away from the person they are talking to makes the call worse, no matter how accurate the suggestion is.
Our solution
We treated the transcript as the product surface rather than a byproduct. Each turn carries its own intent and sentiment label, so the conversation can be read at a glance or drilled into. Guidance lives in a separate rail with a suggested reply, a running compliance checklist and a sentiment trend, none of which move the transcript or steal focus. After the call, the same structure becomes a summary with action items, and scoring runs across every conversation instead of a sample — with anything the model is unsure about routed to a person rather than acted on automatically.
What we built
- 01Real-time transcriptionSpeaker-separated text with word-level timing as the call happens.
- 02Live agent assistSuggested replies and next steps in a side rail that never grabs focus.
- 03Intent detectionEach turn labelled with what the customer is actually asking for.
- 04Sentiment trackingA running read of how the conversation is going, not just a final score.
- 05Automatic summariesEvery call becomes structured notes with extracted action items.
- 06Compliance checklistRequired steps ticked off live, so nothing is missed under pressure.
- 07Topic clusteringRecurring themes grouped automatically from the transcripts themselves.
- 08Coaching scorecardsConsistent scoring across every conversation instead of spot checks.
- 09Playback provenanceAny score or summary line jumps to the moment in the audio behind it.
Desktop
Where the product doesits heaviest work.
The full-width interface, screen by screen.
Mobile
Rebuilt for onehand.
The same information hierarchy, rebuilt for a screen you hold in one hand.
More work
Other projects
(→)Start a project
Want something like Verba for your business?
Tell us what you have in mind and we'll come back with a clear scope, a realistic timeline and an honest opinion on the best way to build it.
Project brief