InsightVoice — Ask the Business by Voice

InsightVoice is a voice-driven business investigation agent designed to make data analysis accessible through natural conversation. Instead of navigating dashboards, writing SQL queries, or manually filtering reports, a user can simply ask a business question by voice. AssemblyAI converts the spoken question into a high-quality transcript, which InsightVoice then routes to specialized analytical tools that investigate real e-commerce data. For the hackathon demo, InsightVoice works with the Olist Brazilian E-commerce dataset and supports three core investigation scenarios: • Revenue Root-Cause — investigates changes in revenue between periods using revenue, order volume, average order value, and category evidence. • Seller Driver — identifies sellers contributing to business performance and surfaces the strongest contributors from the underlying data. • Transaction Anomaly — detects unusual transaction patterns and presents the supporting evidence for further investigation. Rather than returning a generic AI response, InsightVoice produces a structured “Insight Brief” containing a claim, calculated evidence, confidence context, limitations, and a recommended next step. This makes the result easier to verify and more useful for business decision-making. AssemblyAI is the voice layer of the experience. Spoken questions are transcribed before entering the investigation pipeline, allowing users to move naturally from voice to structured analysis. InsightVoice also explores multimodal personal finance capture. Users can add expenses manually, by voice, or by uploading a receipt. Receipt information can be converted into structured transaction data and stored for personal spending analytics. The project combines voice AI, analytical tooling, structured data processing, and evidence-first UX to demonstrate a different way of interacting with business data: ask the business, not the dashboard.