AI
75 notes · 15 categories
Work in progress
I am studying these topics right now and writing the notes as I go. I am also exploring how AI-assisted interview rounds work. If you have a suggestion, email me.
How LLMs WorkTokens, next-token prediction, training, the context window and temperature.5Using LLM APIsMessages, stateless design, cost, streaming and error handling when you call a model from code.5PromptingClear instructions, delimiters, few-shot examples, step-by-step reasoning, chaining and structured output.6HallucinationsWhat hallucination is, the forms it takes, why it happens, and how to reduce and catch it.5Embeddings and Vector SearchEmbeddings, similarity search, vector databases, hybrid search with reranking, and where semantic search fails.5RAGRetrieval-augmented generation: the pipeline, chunking, answer rules, common failures, permissions and when long context is enough.7EvaluationEvals for LLM systems: golden sets, code checks, LLM judges, regression runs and production signals.6Tool CallingTool calling: the request loop, who runs the tool, validation, loop limits and confirming risky actions.5WorkflowsWorkflows your code controls: prompt chaining, routing, parallelization, orchestrator-workers and evaluator-optimizer.6AgentsAgents: the model-driven loop, when to choose one over a workflow, guardrails, precise terms and the trade-off.5MCP and LangGraphMCP as a standard for connecting tools, its three primitives, LangGraph's state graph, and when each is worth the setup.4Safety and SecurityPrompt injection, limiting the damage, secrets, PII, input and output guardrails, and the lethal trifecta.5Production BasicsRunning LLM features in production: latency, cost at scale, reliability, logging and versioning.5Choosing an ApproachChoosing between a better prompt, RAG, fine-tuning, plain code or a hybrid, by naming the gap first.1AI Coding ToolsAI coding tools: the four types, a good workflow, where they fail, the one rule, and how to talk about them in an interview.5