BusinessGuard_AI

Your business's first responder Hacked, scammed or threatened? Get a calm, step-by-step rescue plan by voice or on screen.

Comment

Inspiration

Meet Priya. She runs Saree Haven, a small boutique that sells sarees on Instagram and WhatsApp. One morning she wakes up locked out of her Instagram account. Her customers, her orders and her income are all behind a login she can't use. She has no IT team and no help desk to call, just panic and a pile of conflicting advice online.

Priya is a fictional character, but her story is built from problems small online sellers really describe: hacked accounts, fake payment screenshots and angry customers threatening action. Big companies have security teams for this, and a one-person boutique has nobody.

We wanted to build the first responder Priya never had: something calm that tells her the next step the moment things go wrong, by voice, and then follows her onto the screen.

What it does

BusinessGuard AI is a rescue line for small online sellers. It's a prototype with a simulated Alexa+ experience.

  • Start by voice. Priya says, "Alexa, ask BusinessGuard, my Instagram was hacked." The assistant asks a few quick questions, speaks the first three actions to take, and opens a case with a Case ID.
  • Continue on screen. Opening that Case ID in the web app resumes the full conversation and recovery plan, with a "Continued from Alexa" banner. A QR code lets her pick it up on her phone.
  • Emergency Mode. Step-by-step plans for a hacked Instagram, a compromised WhatsApp, fake payment fraud and a customer legal threat, each with immediate actions, a 24-hour plan and an evidence checklist.
  • Recovery timeline. Steps for the first hour, the first 24 hours and the first week, which she can tick off as she goes.
  • Visible reasoning panel. Shows which checks the agent ran and why it chose a severity level, so the advice isn't a black box.
  • AI Business Advisor. Structured answers with risk level, situation analysis, actions and prevention tips.
  • Payment Checker. Flags risk signals in a payment screenshot and lists what to verify in the bank app. It never claims a guaranteed verdict.
  • Business Health Audit and Protection Score. A 12-question check that produces a maturity level, a score across Legal, Security, Operations and Brand, and the top fixes that raise it.
  • Risk Radar and case management. A quick High, Medium or Low view of cyber, fraud, legal and brand risk, plus every case and its timeline in one place.

How we built it

  • Frontend: React, Vite, TypeScript, Tailwind CSS, Recharts and framer-motion.
  • Backend: Python and FastAPI, with SQLite so the Alexa simulator and the web app share the same case record.
  • MCP server: a self-hosted Model Context Protocol server exposing tools such as create_case, get_case, update_case, analyze_incident, generate_recovery_plan and score_business. The agent calls these tools to reason about each case.
  • Voice: the browser's Web Speech API for speech-to-text and text-to-speech, with typed input as a fallback.
  • AI: the Google Gemini API for live answers, with scripted demo responses that take over automatically if the API fails or hits a free-tier limit.
  • Built with Antigravity, which we used to generate and iterate on the code.

Everything runs locally and for free, with no paid services.

Challenges we ran into

  • Building for free. We had no paid cloud budget, so we added caching and an automatic fallback to demo responses to cope with free-tier limits.
  • Making a simulated Alexa feel real. We built an Echo-style ring with listening, thinking and speaking states, and made the Case ID handoff the centerpiece of the demo.
  • Getting the whole flow to agree on one case. Voice, the database, the MCP tools and the web app all had to read and write the same record. Setup also tripped us up: the backend only started once we ran it from the project root with the right virtual environment.
  • Honesty in a security tool. A tool like this loses trust if it overclaims. We labelled simulated parts and sample data, and we present payment checks as risk signals, not verdicts. The app gives general guidance, not legal or professional advice.

Accomplishments that we're proud of

  • A complete voice-to-screen rescue flow where one case follows the seller from Alexa to the web.
  • A calm, professional interface that stays clear during an emergency.
  • A working app that runs for free and still works when the AI provider fails.

What we learned

  • In a crisis, one clear next step beats a wall of advice.
  • Showing the agent's reasoning builds trust.
  • Designing for failure, with fallbacks and honest labels, makes a demo reliable.

What's next for BusinessGuard AI

  • Publish a real Alexa skill with account linking.
  • Add GST and trademark guidance, a contract reviewer and a compliance calendar.
  • Support Hindi and regional languages.
  • Test with real sellers to improve the recovery plans.

This is a hackathon prototype. Alexa is simulated, data is sample data, and the guidance is general information, not legal or professional advice.

Built With

Share this project:

Updates

Submission history