Andrew Nash

Aurrigo · Staff ADS Engineer

• + Tech lead of 6 engineers

  Across navigation, perception, localization and planning. Code owner of two $1M+ production deployments.

• + Built navigation, perception, localization and calibration for Canada's first medium-speed SAE L4 shuttle

  Transport Canada-authorized. [press]

• + Shipped PyTorch camera+LiDAR 3D detection + tracking at 83.4 mAP @ 20 Hz

  Paired with a classical safety channel for ISO 26262 diverse redundancy.

• + Built a real-time ROS C++ Frenet motion planner

  Replaced legacy PID: 400+ parallel trajectories at 50 Hz, jerk-optimal selection against perception.

• + Cut lateral error 10x (0.5 m to 0.05 m), more than doubled top speed (15 to 35 km/h)

  Custom GTSAM factors with dynamic GPS/LiDAR/IMU sensor fusion. Unlocked tunnel and GPS-denied terminals.

• + Authored Aurrigo's end-to-end Python labelling data pipeline

  Annotation tool + auto-labeller lifting 2D foundation models to 3D LiDAR.

ROS · C++ · PyTorch · CUDA · TensorRT · Git · Computer Vision

Aurrigo · ADS Team Lead, North America

• + Led Aurrigo's first AV simulator from scratch

  1000s of scenarios proving airport-customer safety. 50x speedup vs CARLA, faster than real time.

• + Solo-authored Aurrigo's C++ GTSAM iSAM2 SLAM and mapping

  glim-based GPU factor graph on gtsam_points. GPS-aligned LiDAR maps from all sensors.

• + Doubled top autonomous speed (25 to 50+ km/h)

  27x CUDA VGICP scan-match speedup, multi-LiDAR at 20 Hz, and custom Ackermann + crab-steer kinematic factors, on the same embedded hardware.

ROS · C++ · Python · GTSAM · UE4 · CI/CD · Linux · Bash

Co-op internships · ML, data science and NLP

• Nasdaq (Verafin) Machine Learning Engineer · May – Aug 2021 Lifted gradient-boosting fraud-detection model precision by 6 pp at fixed recall.
• Bank of Canada Data Science Developer · Sept 2020 – Apr 2021 Built anomaly-detection models automating economist data-review workflows.
• IBM NLP Developer · Jan – Apr 2020 Improved enterprise semantic matching by 8% via weighted word embeddings.
• Royal Bank of Canada Full Stack Data Scientist · May – Aug 2019 Built a full-stack semantic search prototype with an NLP backend and web frontend.
• BlackBerry QNX NLP Developer · Sept – Dec 2018 Applied NLP classifiers to automate software-license tagging across QNX repos.

Sage · Real-time D&D AI co-pilot

• Streaming audio to LLM agent at ~1 s end-to-end latency: Deepgram STT, Gemini agent loop, 20+ MCP tools.
• Solo-architected: 98 REST routes, hybrid RAG, Svelte 5 PWA, Docker.

FastAPI · LLM · RAG · MCP · GCP

IGVC · Autonomous vehicle competition

• Trained custom transformer BEV prediction models from Unreal Engine data.
• Design placed 2nd internationally.

ROS · Unreal Engine · C++ · PyTorch write-up →

SR AUV · Autonomous underwater vehicle

• Real-world tested, flying end-to-end autonomously using PPO RL.
• Awarded best 2021 capstone project.

Unity · C# · TensorFlow · TPU · RL write-up →

SpaceX Hyperloop · Pod competition

• Placed 8th at the 2019 Pod Competition. Built full-stack pod telemetry.

C++ · Flask · JS · Protobuf · CAN

Herd's Eye View · AIIDE-2023

Improving game-AI agent learning with collaborative perception. MSc thesis.

Unity · RL · Transformer · PyTorch arXiv →write-up →

MSc Computer Science (Thesis)

Memorial University of Newfoundland. 4.0 GPA. Thesis: Herd's Eye View (AIIDE-2023).

BEng Computer Engineering (Co-op)

Memorial University of Newfoundland. Capstone: SR AUV, best 2021 capstone.