Free, code-first online courses on AI engineering — from first API call to production systems.https://d2apczqz24upf4.cloudfront.net/en-ushttps://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-01/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-01/Make your first Amazon Bedrock Converse API call in boto3: models, inference profiles, account hygiene. Start AIP-C01 prep with Module 1.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-01/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-01/When to fine-tune vs RAG vs prompting, the post-training landscape (SFT, LoRA, DPO, GRPO), and a Qwen3-4B baseline. Free course, Module 1.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-01/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-01/Why LangGraph beats a while-loop agent: nodes, edges, state, and the LangChain 1.0 split. Set up v1.x and run your first StateGraph.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-01/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-01/What agentic AI really is — LLM calls vs workflows vs autonomous agents — plus your first NVIDIA NIM call. Free NCP-AAI course, Module 1.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-01/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-01/What code agents are and why smolagents writes Python instead of JSON — the spectrum of agency, CodeAct, and your first CodeAgent. Free course, Module 1.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-02/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-02/Production prompt engineering on AWS: structured JSON output, Bedrock Prompt Management versioning, and prompt regression tests. AIP-C01 Module 2.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-02/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-02/Curate a tool-calling dataset and format it with apply_chat_template — never hand-roll the prompt. Chat template fine-tuning, Module 2.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-02/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-02/Master LangGraph state: TypedDict schemas, channels, reducers (add_messages, operator.add), MessagesState, and the builder. LangGraph v1.x, Module 2.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-02/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-02/Learn tool calling by building a ReAct agent from scratch, then in LangGraph with NVIDIA NIM. Module 2 of the free NCP-AAI course.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-02/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-02/Watch a smolagents CodeAgent run the ReAct loop step by step: write pandas, hit an error, self-correct, and call final_answer. Module 2.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-03/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-03/Amazon Bedrock model routing, cross-Region inference, ConverseStream, and retry backoff: build the FM integration layer the AIP-C01 exam expects.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-03/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-03/Fine-tune your first LLM with TRL's SFTTrainer and SFTConfig — the correct v1 idioms (processing_class, max_length), hyperparameters, and loss curves.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-03/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-03/Add control flow to a LangGraph agent: conditional edges, the Command primitive, and a bounded self-correcting test-fix cycle. Module 3 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-03/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-03/Master AI agent architecture: chains, routers, supervisors — and when not to build an agent. NCP-AAI Module 3 covers exam Domain 1 (15%).https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-03/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-03/Give a smolagents CodeAgent real powers: build custom tools with @tool and the Tool class, plus web search. smolagents Module 3.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-04/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-04/RAG foundations on AWS: chunking strategies, Titan embeddings, and Amazon S3 Vectors. Build DIY retrieval before managed KBs. AIP-C01 Module 4, Domain 1.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-04/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-04/LoRA freezes the base and trains a tiny B·A adapter — learn r vs alpha, target_modules='all-linear', and ship a few-MB adapter. Free course, Module 4.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-04/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-04/Build a LangGraph agent the v1 way: read-only tools, ToolNode, create_agent (not create_react_agent), and a structured Plan via response_format.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-04/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-04/How AI agents plan, reason, and self-correct: ReAct vs plan-and-execute, reflection loops, reasoning budgets. NCP-AAI Module 4, Domain 5.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-04/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-04/Wire any LLM into smolagents — InferenceClientModel, LiteLLM, or local — with one make_model() factory, plus the real cost truth. Module 4.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-05/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-05/Managed RAG on AWS: Bedrock Knowledge Bases with S3 Vectors, hybrid search, rerankers, citations, and data sync. AIP-C01 Module 5, Domain 1.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-05/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-05/Fit your fine-tune on a free 16 GB Colab T4 with QLoRA: 4-bit NF4, the VRAM budget, OOM fixes — and why it's not a deploy format. Module 5.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-05/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-05/Fan out N tasks in parallel with LangGraph's Send API, aggregate results with operator.add, and gate the fan-in with defer=True. Module 5 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-05/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-05/Give your AI agent memory: LangGraph checkpointers, session persistence, and long-term recall. NCP-AAI Module 5, with a hands-on lab.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-05/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-05/Run a smolagents CodeAgent's untrusted code safely: the threat model, the LocalPythonExecutor's limits, and Docker/E2B sandboxing. Module 5.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-06/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-06/Validate, normalize, and format Bedrock multimodal input: images via Converse, Comprehend, Bedrock Data Automation. AIP-C01 Module 6, Domain 1.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-06/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-06/Replay Anvil's QLoRA faster with Unsloth (measured), and write the same fine-tune as one Axolotl YAML — optional accelerators, no lock-in. Module 6.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-06/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-06/Make your LangGraph agent crash-proof: SqliteSaver, thread_id scoping, get_state, and pending writes. Module 6 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-06/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-06/Build an agentic RAG pipeline with NVIDIA NIM embeddings, Chroma, and reranking — chunking, grounding, and citations for the NCP-AAI exam.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-06/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-06/How smolagents memory works: read agent.memory.steps, continue runs with reset=False, prune context in a step_callback, and replay. Module 6.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-07/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-07/Build an AWS agent with Strands Agents: tool calling, stop conditions, and an MCP server on Lambda. AIP-C01 Module 7, Domain 2.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-07/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-07/How to evaluate a fine-tuned LLM: valid-JSON, function-name & argument metrics, base-vs-finetuned, catastrophic forgetting, LLM-judge bias. Module 7.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-07/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-07/Add two HITL checkpoints to Forge with interrupt() and Command(resume=): plan approval before editing, PR approval before shipping. LangGraph v1.x.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-07/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-07/Design multi-agent systems with supervisors, swarms, and handoffs in LangGraph — plus MCP and A2A — for the NVIDIA NCP-AAI exam (Module 7).https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-07/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-07/Make a smolagents agent reliable and frugal: planning_interval, instructions vs system prompt, and the 6 rules for good agents. Module 7.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-08/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-08/Deploy a multi-agent system on Bedrock AgentCore: Runtime, persistent memory, handoff, and human-in-the-loop. AIP-C01 Module 8, Domain 2.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-08/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-08/DPO vs RLHF: why DPO replaced PPO as the default, building chosen/rejected pairs, the frozen reference model, and tuning beta. Module 8.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-08/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-08/Walk, replay, and fork a failed LangGraph run: get_state_history, update_state, and checkpoint_id in one hands-on Forge debug. Module 8 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-08/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-08/Evaluate AI agents with golden sets, task metrics, and a calibrated LLM-as-judge, then tune with proof — NCP-AAI Domain 3 (13%), hands-on.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-08/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-08/Make a smolagents agent reliable: structured output vs final_answer_checks, the AgentError hierarchy, and self-correction. Module 8.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-09/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-09/Bedrock Guardrails, prompt injection defense, and hallucination control for a production GenAI agent on AWS. AIP-C01 Module 9, Domain 3.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-09/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-09/DPO's family — ORPO, KTO, SimPO: which one for YOUR data? Reference model, pairwise vs binary, single-stage — a decision tree. Module 9.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-09/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-09/Make your LangGraph agent production-durable: durability modes, RetryPolicy on flaky nodes, CachePolicy, and a hardened sandbox. Module 9 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-09/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-09/Add NeMo Guardrails and human-in-the-loop approval to a LangGraph agent. Block prompt injection and ship safe agents — NCP-AAI Module 9.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-09/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-09/Connect a smolagents agent to MCP servers, load tools from the Hub, and publish your own — with the security gates that matter. Module 9.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-10/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-10/GenAI security on AWS: PII redaction with Comprehend, IAM least privilege, audit trails, model cards, responsible AI. AIP-C01 Module 10, Domain 3.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-10/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-10/GRPO tutorial: train a reasoning model with verifiable rewards (RLVR) — PPO without the critic, a deterministic reward_fn, on a free or rented GPU.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-10/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-10/Add cross-thread long-term memory to a LangGraph agent: the Store, namespaces, semantic recall with a free local embedder. Module 10 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-10/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-10/Turn your LangGraph agent into a production API: FastAPI async jobs, Docker, scaling, and cost trade-offs. NCP-AAI Module 10 (Deployment, 13%).https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-10/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-10/Build a smolagents multi-agent system: a manager CodeAgent over a web_researcher sub-agent with managed_agents (orchestrator-workers). Module 10.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-11/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-11/Serverless GenAI deployment on AWS: API Gateway + Lambda, async SQS, EventBridge, CDK, and CI/CD with rollback. AIP-C01 Module 11, Domain 2.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-11/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-11/Merge two fine-tuned models into one with mergekit — no training, no GPU: SLERP vs TIES vs DARE, interference, and re-eval. Model merging, Module 11.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-11/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-11/Master LangGraph streaming: 7 modes, token streaming, custom progress events with get_stream_writer, and astream_events v2. Module 11 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-11/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-11/Run AI agents in production: end-to-end tracing, cost per node, drift detection, and continuous evals. NCP-AAI Module 11, with a Langfuse lab.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-11/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-11/Give a smolagents agent eyes: pass images with run(images=...), pick a VLM, and screenshot pages via step_callbacks. Module 11.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-12/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-12/Cut GenAI costs on AWS: prompt caching, semantic caching, batch inference, and the Flex tier — measured per-ticket. AIP-C01 Module 12, Domain 4.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-12/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-12/Convert model to GGUF, AWQ, or GPTQ: merge_and_unload your fine-tuned adapter, quantize for inference, and pick the format by hardware. Module 12.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-12/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-12/Build multi-agent Forge with LangGraph v1.x subgraphs, a hand-built Supervisor, and Command handoffs — without langgraph-supervisor (broken-forward). Module 12.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-12/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-12/Map the NVIDIA agentic stack — NIM, NeMo, Nemotron, Agent Toolkit — and profile a real agent with NAT. NCP-AAI Module 12 (NVIDIA Platform, 7%).https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-12/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-12/Ground a smolagents agent in a knowledge base: a RetrieverTool (BM25) the agent calls and reformulates, with cited sources. Module 12.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-13/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-13/Evaluate GenAI apps on AWS: Bedrock Evaluations, LLM-as-a-judge, RAG metrics, and a CI regression gate. AIP-C01 Module 13, Domain 5.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-13/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-13/Serve a fine-tuned model two ways: vLLM over the LoRA adapter (--enable-lora) and Ollama over the merged GGUF. Measure latency. Module 13.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-13/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-13/LangGraph's Functional API in practice: @entrypoint, @task, previous, entrypoint.final, and when to choose it over the Graph API. Module 13 of LangGraph in Production.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-13/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-13/Assemble, harden, and ship a multi-agent research assistant: retries, timeouts, production checklist, v1.0 release. NCP-AAI capstone, Module 13.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-13/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-13/Ship a smolagents agent: wrap it in a GradioUI with CSV upload, push it to the Hub as a Space, and run it from the CLI. Module 13.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-14/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-14/GenAI observability on AWS: invocation logs, CloudWatch dashboards, alarms, and a troubleshooting runbook for LLM failures. AIP-C01 Module 14.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-14/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-14/Publish your fine-tuned model to the Hugging Face Hub: push_to_hub, a model card with base_model lineage, and a base-vs-tuned eval. Module 14.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-14/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-14/Deploy a LangGraph agent to production: langgraph.json, langgraph dev, Studio, the SDK (get_client, RemoteGraph), double-texting, and custom auth. Module 14.https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-14/https://d2apczqz24upf4.cloudfront.net/courses/ncp-aai/module-14/What the NCP-AAI exam is really like: format, proctoring, strategy, my debrief after passing, and a free 32-question blueprint-weighted mock exam.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-14/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-14/Trace smolagents with SmolagentsInstrumentor (Langfuse/Phoenix) and score your agent: golden set, LLM-as-judge, TSR, cost/run. Module 14.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-15/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-15/Ship a production GenAI agent on AWS: assemble, harden, and review Relay v1.0 with the Generative AI Lens, then tear it down. AIP-C01 Module 15.https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-15/https://d2apczqz24upf4.cloudfront.net/courses/finetune-prod/module-15/Run the whole fine-tuning pipeline end to end on a free T4 — data, SFT, QLoRA, DPO, GRPO, eval, merge, quantize, publish — and ship Anvil v1.0. Module 15.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-15/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-15/Evaluate a LangGraph agent end-to-end: LangSmith tracing, a 10-issue golden set, openevals judge, agentevals trajectory eval, and a CI regression gate. Ship Forge v1.0.https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-15/https://d2apczqz24upf4.cloudfront.net/courses/smolagents/module-15/Ship a production-grade smolagents agent: run the multi-agent system inside a sandbox (Approach 2), harden it, gate it on evals, and tag v1.0. Module 15.https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-16/https://d2apczqz24upf4.cloudfront.net/courses/aws-genai-pro/module-16/AIP-C01 exam strategy: real format (75 Q/180 min, ordering & matching), a weighted mock exam, and my debrief. AIP-C01 Module 16.https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-16/https://d2apczqz24upf4.cloudfront.net/courses/langgraph-prod/module-16/LangGraph production checklist, 30 mastery questions with explained answers, interview prep, and a real deployment debrief. Module 16 of LangGraph in Production.