You will take research papers, learning algorithms, cloud and agent topologies, and the mathematics underneath them, and turn them into motion, diagrams and charts that an engineer can trust. If you already break down Transformer attention, render linear algebra in Manim, or draw multi-agent systems so the control flow is obvious, this is the work.
What you will do
- Animate theorems, statistics and learning mechanics — Fourier transforms, gradient descent, network dynamics, attention — in 2D and 3D.
- Draw dense, readable architecture: cloud estates, agentic frameworks, model-context protocols, microservice pipelines.
- Read current AI and ML papers and turn the contribution into a short visual thread, a cut, or a chart that can stand next to the source.
- Generate motion from code: Manim first; Blender's Python API, Three.js, Matplotlib / Seaborn, Processing / p5.js where they fit.
- Produce stills and sequences for technical writing, documentation, LinkedIn, X, YouTube, and product launches — infographics, step sequences, cheat-sheets.
- Work with the engineers and writers on the delivery so the picture is accurate, not merely decorative.
What you bring
- Linear algebra, multivariable calculus, probability; Transformers, CNNs, diffusion; agentic workflows you can explain without a slide title doing the work.
- Code-driven graphics — Manim in particular — or WebGL / programmatic 3D (Three.js, p5.js, D3).
- Python for data and custom charts: NumPy, Pandas, Matplotlib, Seaborn, Plotly.
- Enough DevOps literacy to diagram Kubernetes, containers, and AI orchestration without inventing the topology.
- Colour, type, space and hierarchy used to reduce clutter, not to decorate it.
- Figma or Illustrator; After Effects, Blender or Premiere when the cut needs an edit, a voice, or a timed callout on code.
Useful, not required
- A public trail of visual technical breakdowns — X, Substack, YouTube, GitHub, Medium.
- Interactive work on the web: React, WebGL, Canvas.
- LLM prompting and generative pipelines (Midjourney, ComfyUI, Flux, SDXL) used to accelerate assets, not to replace the argument.
What to send
- Three to five samples of technical breakdowns: mathematical animation, learning-system flow, or architecture drawn so it can be checked.
- Repositories or snippets that show programmatic generation — Manim, D3, Matplotlib, or equivalent.