Professional experience

Zero-to-one product ownership for an edge AI mining-safety product

• Built BeltIQ from zero to one, defining the product thesis, the initial feature set, and the market-entry strategy for an edge AI mining-safety product in a market with no incumbent equivalent.
• Ran client requirements gathering directly with site operators and decision-makers, converting operational pain points and safety obligations into a prioritized specification engineering could build against.
• Decomposed an open-ended safety-monitoring problem into iterable delivery chunks, then laid out the technical roadmap and execution plan so every increment shipped something demonstrable rather than deferring value to a single launch.
• Prepared and delivered the sales pitch deck and product narrative used in client and executive conversations, translating edge AI capability into operational and commercial outcomes buyers could evaluate.
• Owned vendor management and procurement for edge compute, cameras, and site-installation hardware, sequencing purchasing against hardware lead times so deployment dates held.
• Drove feature planning and cross-team execution across ML, software, hardware, and field-deployment teams, maintaining a single roadmap all four worked from.
• Acted as the communication layer between client stakeholders and engineering, surfacing tradeoffs, scope changes, and deployment expectations early enough that they stayed decisions rather than becoming escalations.

Production AI systems for safety-critical mining and operational workflows

• Led client-facing AI delivery across product, operations, engineering, and client stakeholders, translating business needs into deployed systems that reduced manual monitoring by 45%.
• Designed on-device multimodal AI systems on NVIDIA Jetson AGX with autonomous perception–decision–action loops running at 30 FPS under strict compute, latency, power, and reliability constraints.
• Architected Video-RAG pipelines using spatio-temporal embeddings, multimodal indexing, and LLM reasoning for contextual search and decision support over live video streams.
• Built real-time and batch inference pipelines using ONNX, TensorRT, CUDA, GStreamer, and NVIDIA Jetson, achieving 2x throughput and 60% latency reduction.
• Automated MLOps workflows with MLflow, AWS EC2, Docker, Git, Jenkins/CI-CD, deployment validation, monitoring, and edge-device runners.
• Developed synthetic data pipelines for rare-event scenarios, reducing data acquisition costs by 50% and accelerating deployment timelines by 25%.

Real-time ML and anomaly detection for infrastructure inspection

• Designed ensemble ML models for real-time classification of structural defects in underwater infrastructure, achieving 90% accuracy and 4x faster processing.
• Built a real-time infrastructure inspection platform integrating live drone feeds, ML predictions, and analytics dashboards, reducing inspection costs by 30%.
• Automated statistical reporting workflows and translated model outputs into stakeholder-facing business insights.

Backend, cloud, and API engineering

• Built scalable REST APIs for multi-source data integration and real-time analytics workflows, improving throughput by 30%.
• Developed and deployed AWS-based platform features focused on backend reliability, performance, and user engagement.
• Worked cross-functionally with product and engineering teams to ship customer-facing improvements.