A personal research knowledge base for Machine Learning and Large Language Models, powered by Claude Code.

This repository organizes research materials (papers, code, documentation) and provides Claude Code skills and commands to facilitate exploration, analysis, and report generation.

mlkb/
├── papers/                # Research papers and paper notes
├── notes/
│   ├── topics/            # In-depth topic notes
│   └── quick-ref/         # Quick reference sheets
├── code-repos/            # Code repository analyses
├── reports/
│   ├── surveys/           # Literature surveys
│   ├── comparisons/       # Comparative analyses
│   └── implementations/   # Implementation guides
├── REFERENCES.md          # Master reference index
├── CLAUDE.md              # Claude Code project config
└── .claude/
    ├── skills/            # Auto-triggering skills
    └── commands/          # Slash commands

/add-ref https://arxiv.org/abs/1706.03762
/add-ref paper "Attention Is All You Need"
/add-ref repo https://github.com/huggingface/transformers

Drop a PDF or provide a URL, then ask:

• "Summarize the key contributions"
• "What methods does this paper use?"
• "Extract the main results"

Provide a repo URL or path:

• "What's the architecture of this repository?"
• "Where is the attention mechanism implemented?"
• "Generate a technical overview"

/note transformers
/note quick-ref pytorch-tips

/survey attention mechanisms
/compare BERT GPT-2 T5

/search attention
/search transformer architecture

These skills activate automatically based on context:

Triggers when analyzing research papers. Extracts:

• Problem statement and contributions
• Methods and algorithms
• Results and benchmarks
• Limitations and future work

Triggers when exploring code repositories. Provides:

• Architecture mapping
• Implementation tracing
• Dependency analysis
• Documentation generation

Triggers when generating reports. Creates:

• Literature surveys
• Comparative analyses
• Implementation guides

papers/YYYY-author-short-title.md
papers/2017-vaswani-attention-is-all-you-need.md

notes/topics/transformer-architecture.md
notes/quick-ref/pytorch-distributed.md

reports/surveys/2026-01-llm-alignment-survey.md
reports/comparisons/attention-mechanisms.md
reports/implementations/lora-fine-tuning-guide.md

code-repos/transformers-analysis.md
code-repos/llama-architecture.md

The knowledge base tracks these reference types in REFERENCES.md:

1. Add papers: /add-ref [paper-url]
2. Analyze each: "Summarize contributions of [paper]"
3. Compare: /compare [paper1] [paper2]
4. Synthesize: /survey [topic]

1. Add repo: /add-ref repo [url]
2. Explore: "What's the structure of this repo?"
3. Trace: "How does the training loop work?"
4. Document: "Generate architecture documentation"

1. Create topic note: /note [topic]
2. Add findings as you research
3. Link to papers and repos
4. Build quick-ref sheets for common patterns

• Keep REFERENCES.md updated as your master index
• Use consistent file naming with dates
• Link related materials across notes
• Tag entries with topics for searchability
• Generate reports incrementally as you learn

Personal research use.