Luka AI: voice-based market research for first-time investors, powered by Vocal Bridge

A voice-based financial assistant that switches charts, builds watchlists, and explains the numbers in plain language, built on Vocal Bridge's Voice for your app surface.

The build wires up eight agent-to-app and two app-to-agent client actions, pulls live prices from Alpha Vantage, and uses Vocal Bridge's post-call processing to email a session summary afterward. It was built by Gekai Liao and Jinglun Gao at the Build future of voice AI hackathon.

The problem: market data is everywhere, understanding is not

A first-time investor opening a brokerage app meets a wall of charts, ratios, and tickers with no one to ask. The data is all there. A PE ratio, an EPS figure, a 52-week high, an analyst target price. What is missing is the part that turns a number into a decision, the explanation of what any of it means and whether it matters.

Forms and dashboards assume you already know the vocabulary. They expect you to know that you should look at an analyst target relative to the current price, or that a rate cut tends to support a growth stock's valuation. A newcomer staring at a candlestick chart does not know what question to type into the search box, because not knowing the question is the whole problem.

What Luka AI built

Luka AI is a voice-based research companion that turns a brokerage screen into a guided conversation. Rather than leave a first-time investor to decode charts alone, Luka drives the interface by voice and narrates the meaning behind the numbers: switching chart views, building watchlists, and surfacing the metrics that actually inform a decision, then explaining why they matter.

The team set out to close the gap between market data and understanding. A new investor rarely lacks information; they lack the context to act on it. Luka's answer is an assistant that watches the same screen the user does, decides what is worth pointing out, and ties a company's fundamentals to the bigger picture (an analyst target sitting above the current price, a PE ratio in context, an interest-rate cut and what it means for a growth stock) in plain language, out loud, as the user follows along.

How Luka AI uses Vocal Bridge

Luka AI is built on Vocal Bridge's Voice for your app surface, the pattern where the voice agent and the UI talk to each other over the same real-time channel. The team implemented eight agent-to-app client actions (switching chart views, adding tickers, navigating pages, surfacing metrics) and two app-to-agent actions so the interface can report state back to Luka. Inspired by plugin-style tooling, they kept the agent configuration as infrastructure-as-code in a config folder so deployment stays reproducible.

import { useAgentActions } from "@vocalbridgeai/react";

const { onAction } = useAgentActions();

// agent -> app: a voice command swaps the portfolio chart
onAction("set_chart_view", ({ symbol, range }) => {
  setChart({ symbol, range }); // e.g. "MSFT", "1M"
});

// agent -> app: add a ticker to the on-screen watchlist
onAction("add_to_watchlist", ({ ticker }) => {
  setWatchlist((w) => [...w, ticker]);
});

Illustrative. The shipped Luka AI build handles a wider set of actions across charts, metrics, and navigation.

The architectural split is clean. Luka owns the portfolio UI, the live market data from Alpha Vantage, the financial reasoning, and the Zapier integration that sends a session summary to the user's Gmail. Vocal Bridge owns the voice channel, the real-time low-latency speech, barge-in, transcript streaming, and the post-call processing that turns the conversation into a follow-up email so the insights are not lost after the call ends.

What's hard about voice here

What Luka pulled off is harder than a chatbot: a spoken market-research session that stays fluid while it pulls live prices mid-sentence and lets the user cut in at any moment. A few things have to be right, and Vocal Bridge handles each:

• Latency has to disappear. A spoken answer that lags feels broken. Prices are fetched live mid-conversation, so the agent has to run a data lookup and keep speaking naturally while it waits. Vocal Bridge's streaming, low-latency speech keeps the reply snappy instead of stalling on the tool call.
• Interruptions have to be graceful. A user will cut in mid-explanation ("sorry to interrupt, can you go back to the main page") and expect the assistant to stop, comply, and pick the thread back up. Vocal Bridge's barge-in handles that turn-taking, so it is not something the team had to build.
• The conversation has to outlast the call. Insight spoken aloud is easy to lose. Vocal Bridge's post-call processing turns the session into a follow-up email, so the takeaways survive after the user hangs up.