Practical AI engineering. From idea to working system.
Design and build AI agents, integrate LLMs with your systems, and evaluate how they perform on the tasks that matter.
Featured Projects
AI agents, persistent memory, video creation and integrations built around practical workflows.
Perceptron Phone
Independent Android prototype. An Android agent that turns a spoken goal into actions in existing apps. It observes the screen, executes an action and checks the result, with checks before consequential actions and before declaring completion.
View Details
Claude Recall
Independent open-source tool. Persistent, project-scoped memory for coding agents that carries decisions and corrections across sessions. Stored rules can be inspected, corrected and retired as a project changes.
View Details
Frame — AI Video Generation & Editing
A local app for generating video, comparing versions and editing footage. Watch three generated-video examples on the project page.
View Details
MCP Server for Intercom
Independent open-source tool. A TypeScript MCP server that gives AI assistants reusable search tools for Intercom conversations and tickets, with customer, date and status filters.
View Details
Autonomous Network Agents
Independent research prototype. Three scheduled AI agents monitor, analyse and expand a simulated Irish 5G network inside a locked-down sandbox. They coordinate through a shared bulletin board and can wake each other when a fault needs attention.
View DetailsCapabilities
Practical support for AI integration, prototyping and evaluation.
AI workflow & tool integration
Connect an assistant to the systems and records a workflow already depends on.
Deliverable: A focused integration with explicit tools, permissions and failure handling.
See MCP Server for IntercomAgent & computer-use prototyping
Explore tasks that require observing an interface, acting and checking the result.
Deliverable: A working proof of concept with action boundaries and recorded behaviour.
See Perceptron PhoneEvaluation & reliability
Separate a plausible response or completed tool call from a verified task outcome.
Deliverable: An evaluation method and report covering representative tasks, failure cases and limits.
See Perceptron PhonePersistent agent context
Carry useful decisions across coding sessions while keeping scope and correction visible.
Deliverable: An inspectable memory workflow and a plan for testing retrieval and stale information.
See Claude RecallA Practical Delivery Approach
Start with a defined problem, build a focused solution and evaluate it against agreed criteria.
Understand the problem
Define the workflow, users, constraints and evidence needed to judge the result.
Build a focused system
Implement a prototype or integration with a specific scope and explicit boundaries.
Evaluate against agreed criteria
Check representative tasks and failure cases. Separate observed results from assumptions.
Document the next decision
Record findings, limitations and what would need validation before wider use.
Engineering Principles
Manifesto for AI-Driven & AI-Assisted Development →
Principles for using AI in software development, including where automation helps and where human judgement matters.
Discuss an AI project
Describe the workflow, integration or technical question you want to explore.
Discuss a project