Research
Notes, papers, and the ideas worth keeping.
This page is the written trail behind the work: what I read, what I learned, and what I want to remember.
Searchable notes
Store articles, observations, and summaries in one place.
Reading list
Books, courses, and papers can all live here.
Thought log
A place for the ideas you want to build on later.
Research notes
From the backendRunning models locally with Ollama + DeepSeek eliminates per-token API costs. Trade-off: need decent hardware (16GB+ RAM for 7B models). Groq provides fast inference via cloud but with rate limits. Best hybrid: local for development/experimentation, Groq for demo/production with caching.
ERC-721: Each token is unique (1:1 NFTs, land deeds, identity). ERC-1155: Multi-token standard — can represent both fungible and non-fungible tokens in a single contract. Used ERC-1155 in Land Trust for batch transfers and gas efficiency. ERC-721 for unique property deeds.
The transformer architecture replaces recurrence with self-attention, enabling parallelization. Key concepts: multi-head attention, positional encoding, layer normalization. The 'Attention Is All You Need' paper introduced this in 2017, and it's now the backbone of GPT, BERT, T5, and every modern LLM.
Learning stack
Use this page for
- Books I have read
- Papers I have reviewed
- Ideas worth revisiting
- Topics I want to study next