Aegis Support

• 

  Aegis

Inspiration

Most AI support demos look good only when every dependency is healthy. Real support teams need agents that can survive outages, avoid hallucinating, and preserve customer trust when systems fail

What it does

Aegis Support is a resilient support operations console. It uses Gemini for natural responses when available, then falls back to verified playbooks when the model path or business tools fail. Judges can run failure scenarios for carrier lookup, model outage, missing policy data, and ticketing failure. The app shows trust score, customer heat, unsafe-answer prevention, recovery timeline, and a human handoff packet

How we built it

React and TypeScript frontend Vite app with server-side development middleware Gemini integration through a backend /api/gemini route Server-side health check through /api/health Deterministic support playbooks for fallback behavior Chaos controls for dependency simulation Responsive operator dashboard

Challenges we ran into

The main challenge was designing an AI agent that still works when its dependencies break. We had to handle Gemini failures, quota issues, missing order data, unavailable policy knowledge, and broken ticketing paths without exposing raw errors to the customer

Another challenge was making the demo easy for judges to understand quickly. That led us to build the guided demo, recovery timeline, trust score, unsafe-answer prevention signal, and executive incident report

Accomplishments that we're proud of

Built a working responsive support console Added Gemini-backed response architecture Added production-style fallback behavior Added guided demo mode for judges Added handoff and incident reporting features

What we learned

We learned that building a useful AI support agent is less about making the chatbot sound smart and more about making it trustworthy when systems fail

The biggest lesson was that production AI needs visible guardrails: fallback routing, trust scoring, escalation packets, and clear uncertainty handling. Aegis became stronger when we stopped hiding failures and made recovery behavior part of the product experience

What's next for Aegis Support

Connect real order, billing, and ticketing systems Add Zendesk, Intercom, and Shopify support integrations Store incidents and handoff packets in a database Add team-level analytics for AI support reliability Add regression tests for agent failure cases

Built With

• api
• css
• custom
• gemini
• github
• lucide
• node.js
• react
• routes
• serverless
• system
• typescript
• ui
• vercel
• vite