Location : Hyderabad, India
About the role : Build real-time detectors for diversion / contraband / bullying by combining statistics, geometry, and lightweight ML on an NVIDIA-accelerated edge box. You’ll turn raw tracks / poses into explainable pattern scores under strict latency budgets.
What you’ll do
- Design zone baselines and lead-lag signals (TLCC, CUSUM / BOCPD, Lomb–Scargle, DTW) to flag coordinated decoys.
- Encode “encirclement,” “follow,” and “throw” with geometry + pose cues; output evidence-rich JSON.
- Write vectorized, well-tested Python (NumPy / PyTorch) that runs
- Collaborate with our DeepStream / TensorRT pipeline to consume atomic events and emit pattern alerts.
Must-have
Strong algorithms + probability; comfort with time series and change detection.Python 3.11, NumPy / Pandas, basic PyTorch; unit tests / benchmarks.Clear thinking about latency, precision / recall, and false-positive tradeoffs.Nice-to-have
TimescaleDB / Postgres, Grafana / Prometheus.Experience with tracking / ReID / pose (even coursework level).Any exposure to DeepStream / TensorRT is a bonus.Why this is cool
You ship detectors that actually protect people, not slideware.You’ll work directly with architects and NVIDIA-stack engineers.Next steps : If interested please email your cover letter and resume to Ashwin at ashwin.jaiswal@safespaceglobal.ai