01
RAG
RAG: retrieval-augmented generation
Having the model answer from external sources: ingestion and indexing, how vector search works, retrieval quality, generation, evaluation and running it in production.
0 entriesView notes
Project 03
LLM Applications and Agents
Learning to build applications on large language models, in two topics: the harness, the part outside the model that lets it carry out tasks as an agent, and RAG, which has the model answer from external sources. Notes are filed by topic, and each starts from a concrete question.
- Status
- Planning
- Started
- September 2026
- Planned tools
- Python · FAISS · pgvector · MCP · LangGraph · RAGAS
This project has not started yet. For now, this page shows the plan and the structure future entries will follow.
What I want to learn
Understand what problem each part of an agent and of RAG solves and how it works underneath, and check the key parts by writing them myself.
Development plan
- 01In progress
The agent loop and a minimal RAG
Write an agent loop without a framework, then plug a minimal RAG with citations into it as a tool.
- 02Planned
Inside vector search
How IVF, PQ and HNSW work and how their parameters trade recall for speed, measured with FAISS, then a simplified index written by hand.
- 03Planned
Context engineering and retrieval quality
Manage the context budget and caching; compare BM25, dense and hybrid retrieval with reranking.
- 04Planned
Tools, MCP, skills and subagents
Build an MCP server, a skill that loads on demand and a read-only subagent.
- 05Planned
Hooks, sandboxes and permissions
Block dangerous actions with a hook, run code in a sandbox, limit tools and skills by role and filter retrieval by permission.
- 06Planned
Memory, durable execution, ingestion and generation
Build per-user memory and resumable execution; compare chunking strategies and check citations.
- 07Planned
Evaluation, observability and production
Evaluate agents with pass@k and pass^k, export traces and make the RAG multi-tenant.
Topics
Notes are filed by topic; each new note goes under the topic it belongs to.
02
Harness
Harness: everything around the model that makes an agent
Everything outside the model: the agent loop, context engineering, tools and MCP, skills, hooks, sandboxes, permissions, memory, durable execution, evaluation and observability.
0 entriesView notes
Overall study roadmap
The full study path for both topics, Harness and RAG: the order to learn them in, how deep to go on each point, common questions and sources.
Read the roadmapEntries
As the project develops, I will add the questions, implementation notes, and test results from each stage here.