The 80% isn't the model.
Eval discipline. Safety surfaces. Observability. The operator who stays disciplined while everything else changes. Two years building production AI systems with Claude Code — eight open-source artifacts public on GitHub.
Founded Steadwise AI in September 2024 to bring agentic AI systems to the Australian SMB market — voice agents on Retell & Anthropic Claude, MCP servers, automation pipelines, evaluation harnesses. Pre-paying-client at the time of writing; capability is real and demonstrable.
Eight public open-source AI artifacts on GitHub. The work that compounded was the boring 80% — eval discipline, safety surfaces, observability — not model-of-the-month chasing. Most-starred repo is AAISAF, an Australian AI Security Assessment Framework covering 87 techniques across 10 tactics with explicit MCP and voice-AI coverage.
Final-semester Bachelor of Business (Finance) at the University of Wollongong, graduating Spring 2026. Currently exploring full-time roles in the Australian AI ecosystem — Forward Deployed, Solutions, Implementation, Customer Engineer, Developer Advocate, AI Solutions Architect.
— Jarrad Bermingham
Eight public artifacts · MIT & CC BY-SA · Five featured below
agent-mcp-frameworkPython framework for multi-agent Model Context Protocol servers. Four orchestration patterns: sequential, parallel, conditional, map-reduce. The substrate for every agentic-workflow build that needs more than one agent.
Python · MIT · Multi-Agent · MCP
View on GitHubaaisafAustralian AI Security Assessment Framework. 87 AI security techniques across 10 tactics, with explicit MCP and voice-AI coverage. Most-starred repo. Authored under Bifrost Labs.
Docs · CC BY-SA 4.0 · AI Security · Governance
View on GitHubbifrost-evalMCP pipeline evaluation toolkit grading agent workflows on accuracy, cost, reliability. Precision@K, Recall@K, F1, MRR. The line between demo and product.
Python · MIT · Evaluation · MCP
View on GitHubclaude-shieldClaude Code lifecycle security hooks. Pattern-matching rule engine for what an agent can’t do. Engineering discipline around guardrails is more valuable than the cleverness of what it can do.
TypeScript · MIT · Safety · Claude Code
View on GitHubrippling-mcp-serverMCP server exposing the Rippling HR/IT/Finance platform to Claude. Employees, departments, leave, groups — surfaced as MCP tools.
TypeScript · MIT · MCP · HR/IT/Finance
View on GitHubCurrently exploring full-time roles in the Australian AI ecosystem. Forward Deployed Engineer, Solutions Engineer, Implementation Engineer, Customer Engineer, Developer Advocate, AI Solutions Architect — the roles where two years of agentic-build experience and a commercial finance background actually compound.
If you’re hiring — or you know a team that values the boring 80% most teams skip — I’d love to talk.