IV. Craft
The instruments
Disciplines
01
AI & ML
- Large language models
- Retrieval-augmented generation
- Agent orchestration hand-built
- Agent workflows in n8n
- Fine-tuning
- Prompt design & evaluation
- Vector search & embeddings
- PyTorch
02
Data
- Python · pandas · NumPy
- scikit-learn
- SQL
- Feature engineering
- Statistics & probability
- Linear algebra
- Visualisation
03
Delivery
- Agile & Scrum
- Scoping & estimation
- Roadmapping
- Stakeholder management
- Client communication
- Requirements translation
- Jira · Confluence
04
Engineering
- FastAPI · REST
- Docker
- Google Cloud
- Git
- APIs & integrations
- Automation pipelines
- Databases
The Method
How I work
-
01
Find the real request
A brief is a translation, not a specification. Before anything gets built I want to know what the person asking would consider a win which is rarely the thing they wrote down.
-
02
Make it measurable early
If a system cannot be evaluated it cannot be improved, and it should not ship. I would rather agree on a crude metric in week one than a perfect one after launch.
-
03
Ship the boring version first
The unglamorous baseline tells you whether the clever approach is worth building. Often it is already good enough, and that is a result rather than a disappointment.
-
04
Keep the timeline honest
Delivery goes wrong quietly, well before anyone admits it. Saying a date has moved is uncomfortable once; discovering it at the deadline is expensive for everyone.
Depth
Where the hours went
Bars show how often each one is in hand, not a score out of ten.
Beside the work
Off the board
You are White
An engine, in your browser. No takebacks.
No moves yet.
I play chess. Badly enough to keep learning, well enough to be annoyed about it. Take a game if you like.
- Rating
- 1350+ · chess.com
- As White
- 1. e4, 2. Nf3 nothing clever
- Playing since
- May 2024
Changing the opponent starts a new game. Ratings are approximate. Engine is Stockfish, GPL-3.0, running in your browser.