Notes

README

3 min read

Notes

A personal knowledge management (PKM) system for professional work in data science, machine learning, and AI. Built in Obsidian.

Purpose

This vault is a structured, long-lived knowledge base — not a note dump. It is organized by epistemic layer and abstraction level: each layer answers a distinct question, from why does this work mathematically? down to why are we solving this problem?

The goal is to accumulate knowledge that compounds over time: timeless theory in foundations, reusable models and algorithms in the modeling layer, proven engineering patterns in the engineering layers, and grounded domain knowledge in applications.

Structure

notes/
├── 00_meta/              # Vault governance: conventions, templates, dashboard, glossary
├── 01_foundations/       # Timeless mathematical and theoretical foundations
├── 02_modeling/          # Model classes and modeling strategies (framework-agnostic)
├── 03_software_engineering/  # Programming, system design, tooling
├── 04_ml_engineering/    # ML productionization: pipelines, deployment, monitoring
├── 05_ai_engineering/    # LLM systems: RAG, fine-tuning, inference, LLMOps
├── 06_applications/      # Domain knowledge: insurance, business context, constraints
├── 07_projects/          # Time-bound execution instances
│   ├── _active/
│   ├── _completed/
│   └── _experimental/
├── 08_reading/           # Structured intake: papers, books, articles, reports
├── 09_logs/              # Operational memory: daily notes, reviews, brain dumps
└── 99_archive/           # Retired content

Layer Guide

LayerQuestion it answersExamples
01_foundationsWhy does this work mathematically?Calculus, linear algebra, probability theory
02_modelingHow do we model this problem?GLMs, neural networks, Bayesian inference
03_software_engineeringHow do we build robust software?System design, Python, Docker, APIs
04_ml_engineeringHow do we productionize ML?Feature stores, MLflow, CI/CD for ML
05_ai_engineeringHow do we operate LLM systems?RAG, fine-tuning pipelines, quantization
06_applicationsWhy are we solving this problem?Insurance pricing, fraud detection
07_projectsWhat are we executing right now?Active work, experiments
08_readingWhat have we learned from external sources?Paper notes, book summaries
09_logsWhat happened today/this week?Daily notes, reflections

Note Types

Notes follow one of six templates (see 00_meta/templates/), matched to their layer and purpose:

Concept note (type: concept) — 01_foundations, 02_modeling

Definition → Intuition → Formal Description → Applications → Trade-offs → Links

Proof note (type: proof) — 01_foundations

Statement → Assumptions → Proof Sketch → Full Proof → Notes / Intuition → Links

Engineering note (type: engineering) — 03_software_engineering, 04_ml_engineering

Purpose → Architecture → Implementation Notes → Trade-offs → References → Links

AI system note (type: ai_system) — 05_ai_engineering

Goal → Architecture → Components → Evaluation → Failure Modes → Cost / Latency → Links

Application note (type: application) — 06_applications

Problem Definition → Domain Context → Data Requirements → Modeling Options → Deployment Constraints → Risks → Links

Project overview (type: project) — 07_projects

Goal → Scope (In / Out) → Deliverables → Data → Modeling → Engineering → Timeline → Links

All notes carry frontmatter with at minimum: layer, type, status (seed / growing / evergreen), tags, and created.

Conventions

Full conventions are in 00_meta/conventions.md. Key rules:

  • File names: lowercase with underscores (chain_rule.md, not ChainRule.md)
  • Top-level structure is stable — changes require updating conventions.md
  • No duplicate concepts — if overlap exists, merge
  • Cross-linking: modeling notes link to foundations; application notes link to modeling and engineering
  • Knowledge flows: Reading → Logs → Projects → Core layers (foundations/modeling/engineering are stable)

Getting Started

Graph View

Backlinks