How do we start using AI in our engineering process?

The path of least resistance is automating repetitive tasks: issue triage, release notes generation, AI-assisted code review, and smart alerts. This creates immediate value without requiring a full process redesign.

What is RAG and when should we use it?

RAG (Retrieval-Augmented Generation) connects an LLM to a specific knowledge base, enabling responses grounded in your company's internal documents rather than just model training. It's ideal for internal chatbots, documentation search, and support assistants.

How much does it cost to run LLMs in production?

It depends on the model and volume. GPT-4o costs approximately $2.50 per 1M input tokens. For high volumes, open-source models (Llama, Mistral) running on your own infrastructure can reduce costs by up to 80%, but require more engineering investment. The right choice depends on the cost-latency-quality tradeoff.

What is an AI Agent and how is it different from a chatbot?

A chatbot answers questions. An AI Agent executes tasks: it accesses systems, makes context-based decisions, calls APIs, processes documents, and acts autonomously within defined boundaries. The difference is between a response and an action.

Claude Code Is Half the Tool Without These 10 Repos

Most developers install Claude Code, get impressed for a week, and then start running into the same walls: it forgets everything between sessions, it wanders off on tangents, it has no idea what your stack looks like unless you explain it every single time.

That’s not a Claude problem. That’s a setup problem.

The open-source community has been quietly solving these issues for months. Here’s what’s worth installing.

1. Claude Mem

github.com/thedotmack/claude-mem

Claude’s memory resets completely every session. Claude Mem gives you a simple layer that persists facts, decisions, and preferences across conversations. It’s not a full knowledge base — it’s closer to a shared notepad that Claude actually reads before answering. Solves the most common complaint people have after the first week of daily use.

2. Awesome Claude Code

github.com/hesreallyhim/awesome-claude-code

A community list of CLAUDE.md files, slash commands, and workflow setups from people who’ve already figured out good configurations. Less a tool, more a library of patterns. Worth reading even if you don’t adopt anything directly — it’s the fastest way to see what’s possible before you go build your own setup from scratch.

3. LightRAG

github.com/hkuds/lightrag

When your codebase or documentation is bigger than what fits in context, LightRAG indexes it and retrieves the relevant pieces at query time. You stop pasting long files manually and start asking questions against a properly indexed knowledge base. The retrieval quality is solid for a project this lightweight. Before building an indexing pipeline, use the Token Counter to check whether your content actually exceeds Claude’s context — you might not need RAG yet.

4. Superpowers

github.com/obra/superpowers

Adds slash commands on top of Claude Code — things like structured planning flows, code review protocols, and debugging routines you can invoke by name. Instead of writing the same long prompts every time, you codify them once and call them like commands. Start here if you’ve ever caught yourself copy-pasting the same instructions into a new session. Before committing anything to a CLAUDE.md, run your prompts through the Prompt Optimizer first.

5. Awesome MCP Servers

github.com/punkpeye/awesome-mcp-servers

A curated list of MCP servers — the connectors that let Claude interact with external tools, APIs, and services. If you’re using anything like n8n-MCP or Obsidian Skills, this is the index for everything else in that ecosystem. Databases, file systems, search engines, dev tools — if someone built an MCP server for it, it’s probably listed here.

6. n8n-MCP

github.com/czlonkowski/n8n-mcp

Connects Claude to n8n through the Model Context Protocol. Instead of describing the workflow you want and then building it yourself, Claude reads your existing workflows, creates new ones, and modifies nodes directly. If n8n is already central to your automation setup, this changes the dynamic pretty significantly.

7. GSD — Get Shit Done

github.com/gsd-build/get-shit-done

Claude has a tendency to explore when you need it to ship. GSD pushes back against that. It structures tasks around concrete outputs and short loops rather than long discussions. If you’ve had Claude spend three messages explaining its approach before writing a single line of code, this is for you.

8. Obsidian Skills

github.com/kepano/obsidian-skills

Lets Claude interact with your Obsidian vault — pull notes, read architecture docs, reference meeting decisions before it starts writing code. Works well for teams that keep their thinking in Obsidian and want Claude to actually use it rather than starting from a blank context every time.

9. UI UX Pro Max

github.com/nextlevelbuilder/ui-ux-pro-max-skill

A set of conventions that gets Claude producing more consistent frontend output — better component naming, tighter structure, less drift between what you described and what gets generated. If you’re using Claude for UI work and keep having to correct the same things, this acts as a standing style guide so you stop repeating yourself.

10. Everything Claude Code

github.com/affaan-m/everything-claude-code

A comprehensive reference for Claude Code patterns — CLAUDE.md templates, workflow recipes, and multi-agent setups. It’s the closest thing to a real manual for Claude Code that exists right now. Covers a lot of ground and gets updated as the tooling evolves.

Where to start

Don’t install all ten at once.

If session memory is your biggest frustration, start with Claude Mem. If you want to see how others have set things up before building your own config, start with Awesome Claude Code. If you’re building on top of MCP, Awesome MCP Servers saves a lot of time finding what already exists.

The rest can wait until you have a clear reason for them. Tools you don’t actually need just add maintenance overhead.

Want this set up for your team?

Installing repos is one thing. Wiring them into a workflow your team will actually use — with persistent memory, n8n integrations, and a RAG layer on your own codebase — is a different kind of work.

That’s exactly what we do at HunterMussel. If you want Claude properly integrated into how your team builds, let’s talk.