Where developers grow together

Resonance — AI Incident Commander

Resonance is a voice-native AI Incident Commander built for EchoSphere 2026 (PS41). It listens to incident calls—spoken live through Agora or entered as text—and transforms what responders say into a structured, continuously updated incident record containing facts, hypotheses, decisions, open questions, actions, and risks, each backed by confidence and evidence. The core problem is that traditional incident-response tools often treat the conversation as one flat transcript, mixing confirmed facts with assumptions and guesses. Resonance keeps these distinctions explicit. Every statement is classified by its certainty, contradictions between responders are automatically detected and surfaced rather than silently resolved, and a fixed 12-point checklist continuously tracks critical unknowns such as customer impact, rollback status, and root cause. A live Clarity Score gives the entire team a simple measure of how well the incident is currently understood. Resonance also keeps humans in control of operational communication. The AI can generate a Slack status update from the confirmed incident state, but the exact message must be explicitly reviewed and approved by a human before it can be sent. This approval requirement is enforced server-side, preventing unauthorized or duplicate automated updates. The system is built with FastAPI and Pydantic on the backend and React + TypeScript on the frontend. It includes real, verified integrations with Google Gemini for structured intelligence extraction, a live Agora Conversational AI voice agent, and real Slack message delivery through the approval gate. It also includes a deterministic, backend-driven Demo Mode for a reliable, judge-ready walkthrough. Test coverage currently includes 141 passing backend tests and 37 passing frontend tests.

Coordinated AI Interview Panel

An adaptive voice interview platform where multiple AI interviewer personas — Technical and Product — share live candidate context instead of operating independently. Built on Agora's Conversational AI Engine for real-time, interruptible voice conversation, with an Orchestrator layer that controls turn-taking between personas and injects shared context so a Product interviewer can challenge a claim the candidate made earlier to the Technical interviewer. Includes a structured final-assessment engine that links every score back to the exact transcript line it's based on, and a clear AI-disclosure notice shown to candidates upfront.Built with Next.js, TypeScript, and Agora's RTC and Conversational AI SDKs.

RELAY — AI Voice Operations Agent

RELAY is a real-time AI voice operations agent built to help customer-support teams resolve customer issues faster while keeping humans in control of sensitive actions. It listens to customers through live voice, understands their intent, verifies order and delivery information, applies business policies, and uses controlled tools to perform actions such as refunds, ticket creation, and escalation. High-risk actions require human approval instead of allowing the AI to act freely. RELAY also supports human takeover, failure recovery, multilingual conversations, live event tracking, and replayable case history. Our goal is to move customer support from AI that only gives answers to AI that can safely complete real operational work.