SupportGenius

Support Genius: Autonomous Support, Orchestrated Intelligence

Inspiration Customer support is broken. Traditional chatbots follow rigid trees that fail on complex edge cases, while human agents are overwhelmed by repetitive volume. We were inspired to build a system that doesn't just "match keywords," but actually reasons. We wanted to move away from the "black box" AI experience and create a platform where you can witness the adversarial process of specialized agents debating the best outcome for a customer.

What it does

Support Genius is an autonomous agent swarm platform. When a ticket arrives, it initiates a high-stakes Adversarial Peer Review:

The Optimist (Customer Advocate): Scours the knowledge base to find every possible way to say "yes" and delight the customer. The Pragmatist (Policy Guardian): Cross-references product catalogs and company policies to ensure every resolution is viable and profitable. These agents debate in a Live Thinking Feed, allowing administrators to see the logic, the disagreement, and the eventual verdict in real-time.

How we built it

The project is built on a modern full-stack architecture:

• Frontend: A premium React dashboard featuring glassmorphism, Framer Motion animations, and a real-time iMessage-style debate log.
• Backend: A Node.js/Express engine coordinating the "Agent Swarm" via sophisticated prompt chaining.
• Data Engine: Elasticsearch is the backbone, powering a product catalog of 100+ items, a directory of 200+ customer profiles, and a high-speed knowledge base.
• Streaming: We used Server-Sent Events (SSE) to stream "thoughts" from the agents directly to the UI, creating a terminal-like experience for AI reasoning.
• Intelligence: Powered by the Gemini 1.5 Pro/Flash models for high-context reasoning.

Challenges we ran into

Coordinating multiple agents to disagree productively was the hardest part. Initially, they tended to agree too quickly. We had to implement a "Devil's Advocate" protocol where the Pragmatist is incentivized to find flaws in the Optimist's plan.

Technical-wise, managing the state of a live-streaming debate across real-time SSE connections required careful synchronization to ensure the UI didn't flicker or lose context as multiple agents "typed" at once.

Accomplishments that we're proud of

We are incredibly proud of the Live Thinking Feed. Seeing the agents argue about a specific product defect or a shipping delay and then watching them arrive at a "human-level" compromise feels like magic. We also managed to seed a robust, realistic environment with hundreds of products and customers to prove the system scales.

What we learned

Building Support Genius taught us that the future of AI isn't a single "God Model," but a swarm of specialists. We learned how to leverage Elasticsearch not just as a database, but as a "long-term memory" that allows agents to recall complex product specs and customer histories in milliseconds.

What's next for Support Genius

• Voice Integration: Letting customers "talk" to the swarm.
• Human-in-the-Loop: Allowing admins to jump into a live debate and steer the agents.
• Automated RCA: Using the agents to provide Root Cause Analysis reports back to the product teams.

Built With

• dark
• express-search-&-data:-elasticsearch-ai-models:-gemini-api-streaming:-server-sent-events-(sse)-design:-glassmorphism
• javascript-frontend:-react
• languages:-typescript
• lucide-react
• mode
• modern
• recharts-backend:-node.js
• tailwind-css
• ui