Neo

Project Description: AI Standup Co-Pilot (Standup Genie)

Fully autonomous, voice-native, async standups powered by real agentic workflows.

Standup Genie is an autonomous agent that replaces synchronous standups with real-time, voice-driven updates. It pulls active Linear issues, launches a Jitsi voice room, speaks with each issue owner using low-latency conversational AI, extracts structured insights via an LLM reasoning service, updates Linear automatically, and posts a daily digest to Slack. It demonstrates an agent that senses (audio), thinks (LLM reasoning), and acts (Linear/Slack), completing an entire organizational workflow end to end.

Judging Criteria: Implementation Reality

1. Working Prototype

The system runs as a full autonomous loop:

Room Provisioning: Backend auto-creates a Jitsi session and exposes signaling tokens.

User Join Detection: Presence events trigger the agent’s activation sequence.

Real-Time Conversational AI: ElevenLabs Conversational AI WebSocket provides unified STT + LLM + TTS at ~150–200ms latency, enabling natural dialog.

Context Injection: Linear issues (title, status, assignee, cycle) are loaded into the agent’s runtime before conversation starts.

Semantic Extraction: A reasoning microservice converts transcript slices into structured fields (status, blockers, ETA, next actions).

Autonomous Linear Writeback: Agent comments, updates estimates, and auto-creates escalation tickets for blockers.

Slack Digest Delivery: PM-ready markdown summaries shipped to Slack channels at the end of each run.

A full demo runs with a single command — triggering Jitsi setup, voice interaction, extraction, writeback, and Slack summary.

2. Technical Complexity & Integration

Custom Jitsi Audio Bridge: Browser client relays mic audio to backend WebSocket and plays back agent audio, using remote track extraction and injection.

Unified Audio Pipeline: 16kHz standardized flow, chunked at 100ms intervals with energy-based VAD for cost and latency optimization.

Conversational Agent Stack: ElevenLabs real-time LLM for reasoning + speech, combined with a secondary extraction LLM for strict schema output.

Linear Two-Way Sync:

Reads cycle issues, states, and metadata for grounding

Writes comments, pushes estimate updates, opens escalation tickets programmatically

Resilience Layer: Reconnect logic for WebSocket drops, Jitsi track recovery, and fallback summary generation if partial audio is lost.

3. Innovation & Creativity

The standup becomes an interactive interview, not status dumping. The agent asks clarifying questions, verifies blockers, challenges vague ETAs, and maintains issue context. It behaves like an always-available project manager with perfect memory and perfect notes.

4. Real-World Impact

Zero facilitation overhead: No human needs to run the standup.

Blockers surfaced instantly: The agent auto-escalates in both Slack and Linear.

High-fidelity context: Updates attach directly to issues, reducing PM drift and slack noise.

Devs stay async: No daily calls, no alignment latency.

5. Theme Alignment

This is a true autonomous agent:

Senses (audio via Jitsi + VAD)

Thinks (LLM reasoning over live transcripts)

Acts (Linear writeback, Slack notifications)

Orchestrates a full workflow with tools and memory
Not a chatbot — a hands-off operational teammate.

Slack and Linear utility modules were partially implemented earlier in the afternoon as supporting components. All agentic logic, voice workflow, Jitsi integration, and ElevenLabs Conversational AI pipeline were built specifically for this project.