Client
Partnership Build
Bebop
AI sales pipeline platform for revenue teams
web appagentb2b
Blackpearl Group
Production AI for New Zealand’s most ambitious companies.
We design, evaluate, and deploy agentic systems inside the workflows where revenue, operations, and decisions actually move.
100+
Systems shipped
CA$13.2M
AI exit
NZ
Operating base
Who We Are
Built by operators from Xero, backed by research cited by Google Brain and AWS, and proven through an international AI exit. Now building for ambitious New Zealand teams.
We put AI inside the workflows where money moves: operations, support, reporting, and decision-making. It ships to production, earns its keep, and starts creating measurable value now.
Applied AI Deployment
Forward deployed AI engineers for companies ready to move past demos.
We embed with your operators, connect AI to the systems where work actually happens, and ship production infrastructure measured by revenue, cost, speed, and decision quality.
Explore deployment→Trusted By
Blackpearl Group (ASX: BPG, NZX: BPG)
Construction Health NZ (CHASNZ)
Forsyth Barr
Buy NZ Made
Pencarrow Private Equity
Milligans Foods
Trans Tasman Media Group
APL Property
Arch Developments
FreshOps
Eco-Index
Our Brands
Wholly owned products - built, launched, and scaled by Levy Street.
Selected Work
Built products, internal agents, and operating systems that leave the lab and survive real users.
Our Research Cited By
Google (Google Brain)
Stanford University
Amazon (AWS)
New York University
Alibaba
NORINCO Group
University of Wisconsin – Madison
Technion – Israel Institute of Technology
Australian National University
Arista Networks
South China University of Technology
Pusan National University
Zillow
Okayama University
Zhuzhou CRRC Times Electric
Xi’an Institute of Modern Control Technology
Pazhou Lab, Guangzhou
Experts In These Systems
Lab agnostic by design. We choose the model, toolchain, and deployment pattern that fits the job.
OpenAI, Anthropic, Google, xAI, Qwen, DeepSeek, and the open model ecosystem all have different strengths. We test them against your workflow, not a vendor preference.
OpenAI
Model lab
CodexGPT-5Responses API
Anthropic
Model lab
Claude CodeClaude CoworkMCP
Model lab
GeminiNano BananaVertex AI
xAI
xAI
Model lab
GrokRealtime reasoningTool use
Qw
Alibaba
Model lab
QwenQwen CoderOpen weights
DS
DeepSeek
Model lab
DeepSeekR1-style reasoningEfficient inference
From the Lab
“Can an LLM write the code that is the model - replacing neural weights with human-readable programs and gradient descent with LLM-guided code editing?”Enter the lab→