- Problem
- Repetitive preparation and iteration can slow creative delivery.
- Why it matters
- Less repetition creates more time for selection, judgement, and finishing.
- Approach
- Run bounded experiments for ideation, rough cuts, asset preparation, and visual iteration.
- Tools
- ComfyUI / Stable Diffusion / Hugging Face / GPT / Gemini
- Current state
- EXPERIMENT
- What was learned
- Clear constraints matter more than long prompts. Human review remains the final quality gate.
- Next iteration
- Document a repeatable workflow with source, output, and review criteria.
THEWORKDOESN'TENDWHENTHEEXPORTFINISHES.
I explore practical ways to reduce repetitive work, organise media, connect tools and build useful systems around creative production.
- MEDIA INTAKE
- ORGANISE
- CREATE
- REVIEW
- DELIVER
- LEARN
Systems Lab index
- Problem
- Manual transfers, naming, and notifications create avoidable production friction.
- Why it matters
- Small reliable automations can protect attention without replacing judgement.
- Approach
- Build small, observable flows and keep every automated step reversible.
- Tools
- n8n / Google Workspace / Notion API / Zapier
- Current state
- LEARNING
- What was learned
- Start with one trigger and one useful output. Logs and failure states are part of the workflow.
- Next iteration
- Connect a verified media handoff and document its failure path.
- Problem
- Cloud-only tools are not always the right fit for private notes or local research.
- Why it matters
- Local models make privacy, latency, and hardware constraints visible.
- Approach
- Compare small local models for tagging, document questions, and research assistance.
- Tools
- LLM Studio / Ollama / Jan
- Current state
- EXPERIMENT
- What was learned
- Hardware limits shape useful model choices. A smaller focused task is easier to evaluate.
- Next iteration
- Create a repeatable evaluation set for one media-organisation task.
- Problem
- A large mixed-media archive becomes hard to browse when naming and metadata drift.
- Why it matters
- A clean archive reduces search time and makes reuse safer.
- Approach
- Use stable public IDs, verified categories, dimensions, and publication states.
- Tools
- Cloudinary / Google Drive / Structured metadata
- Current state
- PROTOTYPE
- What was learned
- Consistency is more useful than complexity. Draft and published states prevent accidental claims.
- Next iteration
- Complete duplicate review and group the strongest assets into verified case studies.