Independent build practice | Ongoing

Applied AI: systems I use every day

A governed system for research, project memory, content, task execution, and multi-agent workflows, used daily rather than built as a demo.

  • AI adoption
  • Agent governance
  • MCP
  • Knowledge operations

Situation

AI tools were creating more output but not automatically creating reliable organisational memory, consistent claims, clean handoffs or safe agent access.

Scope and ownership

  • A daily Obsidian operating system spanning research, evidence, projects, CRM, content and task execution.
  • Multi-agent teams with explicit roles, handoffs, permissions and human decision rights.
  • Public MCP servers that expose selected portfolio context without publishing the private vault.
  • Agent-assisted development used as an independent build practice rather than presented as software-engineering employment.

Operating approach

  1. Make memory durableRoute useful research and decisions into source-of-truth notes instead of leaving them inside disposable chats.
  2. Separate public from privateUse explicit contracts, evidence levels and fail-closed access boundaries for agents and recruiter-facing tools.
  3. Design handoffsGive specialist agents defined inputs, outputs and escalation rules rather than treating a collection of prompts as a team.
  4. Publish inspectable proofTurn the useful parts into repositories, documentation and live interfaces that another person can examine.

What I did

  • Built an Obsidian-based operating system spanning evidence, project memory, research, content and task routing.
  • Created public MCP servers for vault policy, agent handoffs and credential brokering.
  • Developed prompt systems, source-of-truth documents and repeatable production workflows.
  • Used agent-assisted development while preserving human decision rights and explicit public/private boundaries.

Outcome

  • Produced a public portfolio of tested governance-layer MCP servers.
  • Turned AI familiarity into inspectable systems, code, policies and operating artifacts.
  • Created a credible bridge between commercial leadership and technical AI adoption.

What the work taught me

  • The useful unit of AI adoption is a governed workflow, not a model demo.
  • Shared memory becomes valuable only when source quality, permissions and ownership are explicit.
  • Vibe coding can produce credible operating proof when the claims stay honest and the systems remain inspectable.

Recruiter read

Relevant to AI companies that need commercially fluent operators who understand how agents behave inside real workflows, not just how to describe them.

Evidence standard

Public repositories and operating artifacts · demonstrated · reviewed 2026-08-04.

← Back to work