LLMs
Engineering with large language models: orchestration, agents, RAG, and LLMOps.
Phase 1
Advanced LLM Orchestration & Agentic Systems
From linear chains to stateful graphs, tool-calling loops, agent architectures, and multi-agent coordination.
Core Abstractions: Chains vs. State Graphs
Why DAG-shaped chains break down and what LangGraph’s state machines buy you.
- 1.1
Chains vs. State Graphs: Why LangGraph Exists
Linear chains are DAGs. Agents need cycles, shared state, and checkpoints. A look at the abstraction shift.
Updated - 1.2
State vs. Memory: Why Context Windows Aren't Enough
A checkpointer remembers one thread. Agents also need to remember users across sessions and search corpora far larger than any context window. Where each kind of memory lives, and how to wire both into a LangGraph agent.
Tool Calling Mechanics & Execution Loops
What actually happens between the model emitting a tool call and you returning a result.
Agentic Architectures: ReAct, Plan-and-Solve
Interleaved reasoning vs. up-front planning, and when each one fails.
No notes yet.
Multi-Agent Orchestration
Supervisors, hand-offs, and shared state across cooperating agents.
No notes yet.
Phase 2
RAG at Scale
Chunking, hybrid retrieval, re-ranking, evaluation, and serving retrieval in production.
RAG Fundamentals
The indexing and answering pipelines, end to end, and where they break.
Document Processing & Chunking
Parsing PDFs, HTML, and code; structure-aware chunking; metadata worth keeping.
No notes yet.
Hybrid Retrieval
BM25 plus dense vectors, rank fusion, metadata filters, and query rewriting.
No notes yet.
Re-ranking
Cross-encoders, late interaction, and LLM rerankers: precision after recall.
No notes yet.
Evaluating RAG
Retrieval metrics, faithfulness and relevance judges, and building a test set.
No notes yet.
Serving Retrieval in Production
Index freshness, latency and cost budgets, access control, and scaling the index.
No notes yet.
Phase 3
LLMOps
Tracing, evals, gateways, cost control, and deployment pipelines for LLM systems.
Tracing & Observability
Spans for prompts, tool calls, and retrievals; what to log and how to debug a bad answer.
No notes yet.
Evals
Offline test suites, LLM-as-judge, regression gates, and online feedback.
No notes yet.
Gateways & Routing
Provider abstraction, caching, fallbacks, rate limits, and model routing.
No notes yet.
Cost & Latency Control
Token budgets, prompt caching, batching, smaller models, and streaming.
No notes yet.
Deployment Pipelines
Versioning prompts and models, CI for LLM apps, canaries, and rollbacks.
No notes yet.