How snbaskarraj/context_engineering works
Project type and purpose
• This is an educational/demo library of interactive Jupyter notebooks (“context_engineering/”) illustrating four context‐engineering strategies (Write, Select, Compress, Isolate) via the LangGraph framework.
• It is not a traditional web or microservice app but a set of Python data‐flow pipelines driven from notebooks, showcasing how to orchestrate context‐management nodes, memory stores, summarization, and sandboxed execution using LangGraph and external LLM APIs.
File structure
• Top level:
• README.md – project overview and instructions
• requirements.txt – Python dependencies
• context_engineering/ – directory holding four Jupyter notebooks plus utils.py
• context_engineering/utils.py contains common helper code (e.g., state‐graph builders, memory store wrappers).
• No web server, no frontend UI; notebooks serve as the “UI.”
Key insights from README
• Four distinct pipelines are implemented as notebooks using LangGraph:
1. Write Context (state persistence)
2. Select Context (relevance retrieval)
3. Compress Context (summarization)
4. Isolate Context (multi-agent/sandbox)
• Dependencies: Python ≥3.9, uv package manager, Deno (for sandboxing), LangGraph, OpenAI/Anthropic APIs.