User-controlled paper order — the reviewed terms and local simulation receipt.
Voice-first research
Portfolio context — the sample exposure view, including its data-coverage caveat.
See both sides — the research result with supporting evidence and countercase.

Inspiration

Stock research can mean moving between company reports, market information, and portfolio screens before an investor can understand the case for a company or how it relates to what they already own. We wanted to explore a more natural way to bring those steps together: ask a question, hear a balanced answer, inspect its evidence, and decide what to do next.

Our goal was to make research easier to approach without turning a voice assistant into an authority that tells people what to buy.

What we built

Market Partner is a voice-first research and portfolio companion for self-directed investors. In the demo, a user can:

Ask for the strongest case for and against a company, and see the supporting evidence and open questions.
Ask about their exposure to that company and get an answer grounded in a sample portfolio.
State exact terms for a paper order, review them on screen, and separately confirm a local simulation.

The experience connects research, portfolio context, and a possible next step in one conversation. The assistant does not choose a stock, recommend a trade, or place a live brokerage order.

How we built it

The browser app uses React and TypeScript for the voice-and-touch interface, with browser speech APIs and text input as a fallback. A Python FastAPI service hosts the conversation workflow and a self-hosted Model Context Protocol (MCP) server. The browser conversation host discovers and calls tools over Streamable HTTP. The research workflow uses LangGraph and Sarvam AI; deterministic application logic handles portfolio exposure calculations and paper-order validation.

The Alexa+ style interface is a browser simulator backed by our MCP server. It is not a native Alexa+ client and does not connect to an Alexa+ device or private Alexa+ runtime.

Challenges and what we learned

A useful voice experience needs more than spoken answers. People need to see what the assistant used, where the uncertainty is, how an answer relates to their portfolio, and exactly what they are confirming. We designed the demo to keep evidence visible and to separate an order preview from confirmation.

We also learned to keep the boundary between AI analysis and deterministic application logic clear: the research workflow organizes and summarizes the sample evidence, while calculations and order-state checks are handled by application code.

Demo boundaries

The demo uses fictional companies and synthetic sample portfolio data. It has no live quotes or live brokerage connection; orders are local paper simulations. These limits are displayed in the interface so viewers can distinguish the prototype from a production investing service.