Watcher Your AI Incident Commander | Agora EchoSphere 2026 | Team STARK

Watcher — Your AI Incident Commander Built for Agora EchoSphere 2026 by Team STARK When production breaks, the first ten minutes are chaos: people scrambling in a call, no one writing anything down, decisions made and forgotten, and the postmortem written from memory hours later. Watcher sits in that call as a real AI incident commander — not a chatbot, not a scribe — listening live and turning the conversation into structured, actionable state as it happens. What it does Detects and triages incidents automatically from live service logs — deterministic scoring first, AI-based triage (Gemini) only where it adds judgment, never guessing at root cause on its own. Joins the actual incident call via Agora's Conversational AI Engine and listens in real time, extracting facts, hypotheses, decisions, action items, and open conflicts as people talk — no one has to remember to write anything down. Speaks up only when it matters: flags contradictions the moment they happen, nudges on missing information, catches when action items have no owner, and tracks a live coordination-health score that surfaces a status recap if the call is visibly going off the rails — not on a timer, on actual signal. Resolves who's responsible for what by cross-referencing a real team directory first, falling back to reading what people actually said and did on the call — never guessing at someone's role. Wraps up with a spoken recap and a full post-call summary, then — and only with a human clicking approve — opens a Slack approval flow and files a Jira ticket. No autonomous actions ever reach a real external system. Why this matters Every AI-driven decision in Watcher is advisory until a human signs off. Two separate, private approval gates (one for the incident, one for filing the ticket) mean nothing gets escalated, paged, or ticketed without someone explicitly saying yes — augmenting the incident commander, never replacing their judgment. Built with Agora Conversational AI Engine (RTC, RTM, native MCP tool-calling) for the live voice layer; FastAPI + SQLAlchemy backend; Google Gemini for AI triage and call-role inference; Slack + Jira for human-gated external actions.