SELECTED WORK

PROJECTS

AI beyond the output.

I approach these projects by understanding how the model works, designing the system around it and checking whether it helps someone do something useful. ML4Schools is my flagship work, bringing that approach into AI education.

Model foundations
Start with the data, the learning process and how to evaluate a result. ML4Schools makes those ideas tangible through model training and hands-on experiments.
Agent engineering
Give models useful tools and a clear workflow. The content harness combines MCP integrations, shared context, validation and human review.
Real-world delivery
Connect the technical system to the person using it: classroom workflows, reviewable content and experiments with voice and gesture control.

01 / FLAGSHIP PROJECT

ML4Schools

FOUNDER & ENGINEER · NINNOVATE

Raising the floor of AI literacy through project-based learning.

AI and machine learning education for Australian schools, connecting practical projects and browser-based labs with the concepts behind them.

Explore ml4schools.com.au

Why I built it

My own path from high school through my AI and data science degree shaped this project. I felt there was a better way to teach how AI works through project-based learning, drawing on the way I had learned it myself.

I wanted to bring that experience to more students and help raise the floor of AI literacy.

The learning rationale

The aim is to make AI something students can investigate: work with data, train a model, test its behaviour and decide when to trust a prediction.

Prepared datasets, sequenced activities and models that run in the browser let teachers focus on the learning. Each experiment connects practical decisions to the concepts behind them.

What I bring to it

As founder and engineer, I work across the student, teacher and administration experiences: browser-based model training, shared content contracts and the workflows that connect them.

I also build the agent tooling behind content creation, with structured generation, validation and human review.

Explore the content generation harness

02 / AGENT INFRASTRUCTURE

Content generation
harness and agent

MCP · TypeScript · Agent systems

I built a content-authoring harness for ML4Schools that connects AI agents to the platform through the Model Context Protocol (MCP).

It turns a content brief into structured lessons, custom interactive elements and staged assets that a person can inspect, refine and approve.

Why I built it

Project-based AI education needs more than written explanations. It needs activities students can explore, datasets they can work with and models they can test.

I built the harness to bring agents into that authoring workflow, giving them the context and tools to create material that fits the learning experience and the platform.

The problem it solves

A generated lesson still needs the right structure, working asset references and interactive elements that integrate with the host application.

The harness connects those steps: shared context, typed tool operations, asset staging and validation feedback. Agents can correct their work before handing it back for human review, with approval and saving kept under administrator control.

Technical skills demonstrated

MCP & TypeScript: designing tool interfaces and shared contracts between agents and the application.

Agent systems: versioned context, durable workspaces and restart reconciliation.

Validation & integration: actionable error feedback, sandboxed HTML activities, dataset/model staging and human review boundaries.

Explore the MCP and custom elements

Custom learning elements

The agent can author self-contained HTML activities: parameter explorers, sorting tasks, step-by-step mechanisms and annotated diagrams.

Datasets and model assets can be staged alongside the content, supporting hands-on machine-learning labs as well as written material.

Tools with clear contracts

The MCP exposes focused operations for reading the build context, adding content, staging dependencies and checking the workspace.

Typed contracts and validation give the agent concrete feedback to work from. It can revise a result instead of treating its first output as finished.

Review stays with people

Generated assets are staged for inspection. HTML previews run in a sandbox, with checks for content contracts, host integration and design requirements.

The agent does not publish directly. Approval and saving remain administrator actions, keeping generated work separate from live content until reviewed.

03 / MULTIMODAL INTERFACES

Agentic gesture controls

A multimodal control layer that translates voice and hand gestures into structured commands for coding agents.

Voice · Gesture tracking · Coding agents

How it works

Voice captures the intent of a task, while a gesture-tracking harness supplies spatial control. The control layer combines those inputs into structured commands for coding agents such as Codex and Claude Code.

The coding agents then drive the connected tools and functions. The project explores how voice and hand movement can work together as an interface for agent-assisted software development.

Gesture input controlling the agent interface. Open video

Let’s build something useful.

Applied AI, thoughtful products and the systems behind them.

nicholasschlederer@gmail.com

PROJECTS

AI beyond the output.

Model foundations

Agent engineering

Real-world delivery

ML4Schools

Why I built it

The learning rationale

What I bring to it

Content generation harness and agent

Why I built it

The problem it solves

Technical skills demonstrated

Custom learning elements

Tools with clear contracts

Review stays with people

Agentic gesture controls

How it works

Let’s build something useful.

Content generation
harness and agent