LEARNING

Tools & Frameworks

Created 2 May 2025
learningtoolsframeworkspracticaldevelopment

Tools & Frameworks

What is it?

The ecosystem of libraries, platforms, and developer tools used to build, train, deploy, and interact with AI systems.


Development Frameworks

PyTorch

  • What: Deep learning framework (Meta)
  • Use: Research, training, model development
  • Why it dominates: Pythonic, flexible, great debugging, massive community
  • Website: pytorch.org

Hugging Face Transformers

  • What: Library of pre-trained models + tools to use/fine-tune them
  • Use: Load any model in 3 lines of code, fine-tune, evaluate
  • Why it matters: Democratised access to AI models. The “npm of AI.”
  • Website: huggingface.co

LangChain

  • What: Framework for building LLM-powered applications
  • Use: Chains, agents, RAG, tool use, memory
  • Note: Widely used but sometimes criticised for complexity/abstraction
  • Website: langchain.com

LlamaIndex

  • What: Data framework for LLM applications
  • Use: RAG, document indexing, retrieval, knowledge bases
  • Website: llamaindex.ai

Ollama

  • What: Run LLMs locally with one command
  • Use: Local inference, privacy, experimentation
  • Why it matters: Makes running open models trivially easy
  • Website: ollama.com

AI-Assisted Coding

ToolTypeNotes
Claude CodeTerminal agentReads/writes code, runs tests, autonomous
GitHub CopilotIDE autocompleteInline suggestions, chat
CursorAI-native IDEFork of VS Code with deep AI integration
Windsurf (Codeium)IDEAI-native development environment
AiderTerminal toolAI pair programming in terminal
ContinueIDE extensionOpen-source AI coding assistant

Vector Databases

ToolTypeBest for
PineconeManaged SaaSProduction, no-ops
WeaviateOpen-sourceHybrid search, flexibility
ChromaDBOpen-sourceSimplicity, Python-native
pgvectorPostgreSQL ext.Already using Postgres
QdrantOpen-sourcePerformance, filtering
MilvusOpen-sourceLarge-scale production

Orchestration & Deployment

ToolPurpose
LangSmithLLM observability and debugging
Weights & BiasesExperiment tracking
vLLMFast LLM inference server
TGI (HuggingFace)Text generation inference
Vercel AI SDKBuild AI apps with Next.js/SvelteKit
ReplicateRun models via API

Model Context Protocol (MCP)

Anthropic’s open standard for connecting AI to tools:

  • Universal interface for tool/data access
  • Servers expose capabilities, clients (Claude, etc.) use them
  • Growing ecosystem of integrations
  • Relevant to agent development

What to Learn First

If starting from scratch:

  1. Python basics → essential for AI/ML
  2. Hugging Face → use pre-trained models immediately
  3. Ollama → run models locally, experiment freely
  4. LangChain or LlamaIndex → build RAG apps
  5. An AI coding tool → accelerate your own development

Resources

  • Hugging Face course (free)
  • LangChain documentation & tutorials
  • Ollama model library
  • RAG & Retrieval — Core use case for these tools
  • AI Agents — Where orchestration frameworks shine
enes