How eyelevelai/code-samples works
Project’s nature and purpose
• This is a *monorepo* of assorted code samples: Jupyter notebooks (“notebooks/”), standalone Python demos (“python/”), TypeScript demos (“typescript/”), RAG-battle experiments (“rag-battles/”), plus two larger subprojects under “projects/”:
• A Python-based Relativity database uploader (relativity-database-uploader/)
• A full-stack search & completion demo in both Next.js (search-and-completion-next/) and a Node.js service (search-and-completion-node/)
• There is no single running system; instead each folder represents a self-contained sample or micro-application.
File structure to spot architectural patterns
• Top-level groupings:
• notebooks/: data science and RAG-guide notebooks
• projects/relativity-database-uploader: Python CLI + DB modules + Relativity API integration
• projects/search-and-completion-next: Next.js 13 app with React pages + API routes + LLM client services
• projects/search-and-completion-node: Node.js CLI/server invoking LLM services
• python/*, typescript/*: simple demos illustrating environment-based LLM calls
• rag-battles/*: multi-stage RAG experiment pipelines (data ingestion, chunking, retrieval, LLM calls, results/upload)
• Look for sub-components like API layers, service modules, config files, external calls.
Extract insights from the README
• The root README is minimal (“# code-samples”)—it simply brands the repo.
• Each subproject often contains its own README pointing out stack, usage, and environment variables.
Detail without overthinking
• *all* high-level components, even if they only live in separate folders
• Don’t worry about inter-sample interactions—each is standalone
• How data flows within each sample: entrypoint → modules → external systems → output