Peter Hollows

Professional Profile

Twenty years engineering and researching AI systems, from hardware-integrated simulation to large-scale ML deployment. Currently Senior Manager, AI & Data at a Big 4 consulting firm, with an active personal practice in reinforcement learning, probabilistic modelling, and GPU systems. Much of my research has been in sensitive client environments; published writing is on the blog.

Skills

• GPU computing: custom CUDA kernels, JAX, SM occupancy and register pressure optimisation, CPU/GPU pipeline overlap
• Reinforcement learning: DDPG, multi-agent systems, RLHF, MAML, LoRA fine-tuning
• LLMs: RAG systems, evaluation frameworks, routing, latency/cost optimisation
• Machine learning: deep learning, GNNs, probabilistic modelling, pairwise ranking
• ML systems: training pipelines, large-scale data ingestion, model deployment
• Data engineering: PySpark, TimescaleDB, ZFS, Python, Rust
• Cloud and infrastructure: AWS, Azure, Linux, HPC, Docker
• Web and API: FastAPI, Redis Streams, PostgreSQL, vector databases

Certifications

• Stanford: AI Professional Certification, Reinforcement Learning, NLP, Graph Neural Networks, Meta Learning
• MIT: Data to Insights, Probability Theory, Manufacturing
• Deeplearning.ai: Full 5-course AI Specialisation

Side projects

• Custom CUDA kernel for secp256k1 EC multiplication achieving 166M keys/sec on an RTX 3090 (500x CPU), with SM occupancy optimisation and a double-buffer CPU/GPU pipeline — see Keycarver
• Self-hosted tick data pipeline handling billions of rows for ML training, orchestrated with Python and Rust, optimised on a local ZFS/TimescaleDB stack
• Sports prediction models for AFL, greyhounds, and tennis: live odds scraping, pairwise ranking models, and execution engines
• Reinforcement learning: swarm-based multi-agent environments (DDPG variants), student-teacher LLMs with early RLHF, attention mechanisms for GNNs, MAML/LoRA experiments — see Ultimately the Survivors Do Not Prevail
• Ongoing: applying ML and RL techniques to a hard cryptographic problem using JAX on GPU — write-up in progress
• Personal GitHub: captainpete

Career

In consulting, I’ve worked on 10+ large client engagements over 15 years across banking, mining, government, and health. Additional details available on request.

Pre-consulting work spanned real-time flight simulation with hardware integration (Pacific Simulators), retail PoS and online systems (Circle Software), founding a Rails consultancy delivering rapid prototypes for US clients (White Hat Consulting), and full-stack web and iOS development at Trike Apps, where products included early versions of Bellroy and Radiopaedia.

Details

Australian citizen, also hold New Zealand citizenship. Based on the Sunshine Coast (it’s fantastic); remote-first, available to travel on occasion, open to relocation if the role warrants it. Enquiries: recruiters@dojo7.com