11 · Life Plan
Tasks
All aligned with the master plan. Things I'm getting done but not necessarily everyday. Finer than the Master Plan's artifact backlog. Every item here is something you can ask me about. Each varies in difficulty and length. All published here for accountability.
Lifetime
- Python AI Experiment TemplateopenAI
A reusable repo structure for AI experiments.
Includes:
- notebooks
- scripts
- data folder
- results folder
- README
- environment file
- experiment log template
Done means:
- template repo exists
- one sample experiment is included
- setup instructions work
Priority: High
- Classical ML Basics RepoopenAI
A repository containing basic ML experiments.
Projects:
- linear regression
- logistic regression
- classification metrics
- train/test split
- confusion matrix
- model comparison
Done means:
- at least 5 notebooks/scripts exist
- metrics are explained
- README explains what was learned
Priority: High
- Deep Learning LabopenAI
PyTorch and/or TensorFlow/Keras notebooks covering tensors, training loops, image classification, text classification, and model saving.
Done means:
- basic neural network trained
- results documented
- overfitting discussed
- model saved and loaded
- README explains workflow
Priority: High after ML basics
- LLM API PlaygroundopenAI
A repo for experimenting with model APIs, prompts, structured outputs, latency, cost, model comparison, and failure cases.
Done means:
- API calls work
- structured output example exists
- cost/latency notes exist
- prompt examples are versioned
- failure cases are documented Priority: Very high
- Source-Grounded RAG DocumentopenAI
Assistant A document assistant that answers only from uploaded or indexed sources and provides citations.
Done means:
- document ingestion works
- embeddings/indexing work
- retrieval works
- answer generation works
- citations/source references appear
- retrieval failures are documented
- README explains limitations
Priority: Very high
- Ollama Local Model LabopenAI
A lab for running local models, testing local inference, embeddings, structured outputs, and local RAG.
Done means:
- Ollama setup notes exist
- at least two local models tested
- latency/memory notes exist
- local embedding example exists
- local RAG demo exists
- cloud vs local comparison exists
Priority: Medium-high
- Agent Workflow LabopenAI
A repo for tool-calling workflows, agents, human approval, logs, and failure analysis.
Done means:
- at least one workflow exists
- at least one tool-calling agent exists
- logs/traces are recorded
- human approval point exists
- failure cases documented
Priority: Medium-high
- AI Evaluation LabopenAI
A repo for prompt tests, RAG evals, model comparisons, human grading, LLM-as-judge experiments, and failure taxonomies.
Done means:
- eval dataset exists
- at least one model comparison exists
- RAG evaluation exists
- failure taxonomy exists
- before/after improvement report exists
Priority: High
- AI Product Case StudyopenAI
A full case study of one AI system.
It should cover:
- problem
- users
- model choice
- data
- prompts
- retrieval
- evals
- failures
- risks
- limitations
- improvements
Done means:
- published on website or GitHub
- includes architecture diagram
- includes examples and failure cases
- does not overclaim
Priority: High once one AI project exists