Bridging the gap between technical meeting transcripts and actionable Agile execution. LUME transforms raw audio data into structured sprint intelligence, ensuring no technical decision or project risk goes unmonitored.
Experience the LUME interface in action:
NLP-driven decision mining from live transcripts.
Drag-and-drop task orchestration with real-time state sync.
Predictive visualization of project bottlenecks and risks.
High-fidelity simulated mode for offline strategy sessions.
LUME is built for resilience. The system automatically toggles between Live Mode (powered by Gemini 2.0 Flash) and Simulated Mode (High-Fidelity Mock Fallback). This ensures the dashboard remains functional even in strictly air-gapped environments or during API outages.
- Live Mode: Real-time analysis via
useMeetingAIstreaming to Gemini. - Simulated Mode: Automated failover to industrial-grade mock datasets for planning sessions.
Decision mining at the edge. LUME uses advanced NLP to scan transcripts for specific intent markers. It identifies:
- Action Items: Automatic extraction of owners and priorities.
- Technical Decisions: Mining for architectural consensus.
- Risk Indicators: Flagging potential delays before they impact the sprint.
A dynamic orchestration layer for task management. All action items extracted from the meeting are instantly populated on the board.
- Real-time Sync: State updates across all observers.
- Categorization: Automatic tagging (Technical, Product, Design, QA).
A predictive visualization engine. By analyzing the impact and reason for identified risks, LUME generates a heatmap that allows project managers to:
- Identify single points of failure.
- Visualize project bottlenecks.
- Click-to-source: Jump directly to the transcript line that triggered the risk.
Engineered for speed and precision:
- React.memo: Aggressive memoization of complex UI components (Heatmaps, Timelines) to prevent unnecessary re-renders.
-
$O(1)$ Search Indexing: An inverted index is pre-calculated from the transcript, allowing sub-millisecond keyword highlighting across thousands of lines.
- AI Core: Google Gemini 2.0 Flash integration via
useMeetingAI. - Frontend: React 19 (Concurrent Mode) & Tailwind CSS 4.
- State: Centralized orchestration via React hooks.
- Testing: Vitest for high-velocity validation of simulation states.
graph TD
A[Meeting Input] -->|Transcript Stream| B(Gemini 2.0 Flash)
B -->|Structured JSON| C{State Orchestrator}
C -->|Simulated Fallback| D[Mock Data Engine]
C -->|Live Data| E[UI Components]
E --> F[Transcript Timeline]
E --> G[Sprint Planner]
E --> H[Risk Heatmap]
The LUME test suite, powered by Vitest, ensures 100% reliability of the simulation engine.
- State Transitions: Validating the handoff between Live and Simulated modes.
- Data Integrity: Ensuring AI-generated JSON conforms to strict technical schemas.
-
UI Resilience: Regression testing of the
$O(1)$ search index logic.
Distributed under the MIT License. See LICENSE for more information.