GIF
flight timelapse
GIF
individual aircraft selection
GIF
gemini aircraft analysis
GIF
preset zoom settings
emergency detected from abnormal altitude reversal
emergency detected from abnormal speed loss
emergency detected after significant telemetry gap

Syren

Inspiration

Air traffic tracking tools offer a window into the sky, but traditional radar visualizations leave non-experts guessing about squawk anomalies, emergency statuses, and flight behaviors. We set out to bridge the gap between complex ADS-B telemetry and actionable intelligence. Syren was born from the idea that anyone should be able to view real-time and historical flight data paired with instantaneous, context-aware AI explanations.

What it does

Syren is a real-time and historical flight tracking platform with an integrated AI copilot:

Live & Replay Mapping: Stream live airspace traffic or replay target historical dates using high-efficiency binary search snapshotting.
Anomaly Detection: Highlight emergency squawk codes and flight irregularities automatically using visual alerts.
AI Copilot (Gemini): Select any aircraft to trigger an instantaneous analysis that breaks down state metrics, flight trajectory, and anomalies into structured insights.
Data Export: Export airspace snapshots and historical flight traces for custom analytical workflows.

How we built it

Frontend: Built with React and TypeScript (src/App.tsx), utilizing Deck.GL for high-performance canvas rendering of active aircraft icons and trajectory traces.
Backend Adapter: Powered by Python, featuring a live data adapter (backend/live_adapter.py) that filters, normalizes, and flags stale position data.
Replay Engine: Implemented in backend/replay.py and backend/history.py to dynamically fetch archive datasets from GitHub releases and compute instantaneous airspace states using binary search (bisect) over timestamped arrays.
AI Analysis Pipeline: Integrates Gemini 3.8 Flash (backend/gemini.py) via a background thread job-polling architecture to analyze JSON flight states without blocking client requests.

Challenges we ran into

Strict Prompt Constraints: Engineering the Gemini prompt to return structured plain-text insights while handling missing parameters cleanly without outputting invalid markdown formatting.
Handling High-Volume Telemetry: Sorting and filtering large historical datasets required optimizing memory usage and implementing windowed lookups (GONE_AFTER_S) to render smooth replays without lag.
Asynchronous Polling: Preventing HTTP timeouts during AI generation required building an asynchronous job queue with progress-polling endpoints on the backend.

Accomplishments that we're proud of

Seamlessly rendering thousands of airborne aircraft concurrently on Deck.GL with custom visual alert triggers.
Achieving near-instantaneous flight state reconstruction from raw historical archives using optimized binary search algorithms.
Successfully stitching live telemetry, replay history, and multi-threaded AI analysis into a single responsive dashboard.

What we learned

Strategies for constraining LLM outputs to rigid structural formats while preserving natural-language explanatory power.
Techniques for managing high-frequency spatial datasets and real-time state normalization in Python backend pipelines.
Practical pattern design for decoupling heavy background tasks from frontend client UI loops.

What's next for Syren

Ground Aircraft Support: Incorporate ground-tracked aircraft and airport taxi operations into the map interface.
Enhanced Export Feedback: Hook intermediate progress tracking directly to the export UI for granular progress feedback.
Multi-Aircraft Comparison: Allow users to select multiple aircraft simultaneously to evaluate airspace congestion, formation flights, and proximity warnings.