AI is reinventing software. Join the new wave of development.

• Context engineering
• Spec-driven development
• Code review
• Test-driven development
• Harness authoring
• Agent orchestration
• Governance

1. L1Fundamentals and the New Paradigm+

   LLMs, next-token prediction, training, reasoning models, agents, tools, the agent loop, skills, memory, the harness, the context window, Claude Code, agent limitations, the SDLC, and the director role.

2. L2The harness: shaping and extending the agent+

   The Claude Code harness, CLAUDE.md, rules, skills, slash commands, subagents, hooks, permissions, MCP, and the Claude Agent SDK.

3. L3Context engineering and development by specification+

   The context window, context rot, /context, /compact, /clear, write, select, compress, isolate, specification-driven development, spec sections, the /spec command, plans, and verification.

4. L4Generation from spec, code review, testing, and debugging+

   Code generation, planning, gates, iteration, autonomy, code review, blast radius, review commands, reviewer sub-agents, red-green-refactor, test coverage, characterization tests, validation gates, debugging, and regression tests.

5. L5Harness development+

   Selection of primitives, Skills, sub-agents, scope and precedence, built-in sub-agents, plugins, hooks, events, matchers, exit codes, MCP servers, CLI vs. MCP, and the Stop-hook gate.

6. L6Orchestration and governance+

   Orchestration, context isolation, file isolation, git worktrees, fan-out with sub-agents, agent teams, fan-out with worktrees, headless fan-out, dynamic workflows, governance, invariants, security hooks, deny rules, sandboxing, cost control, and team standards.

• Transformer architecture
• Attention
• Embeddings
• LoRA Fine-tuning
• Model routing
• LLM evaluation
• Data sovereignty

1. L1Deep Learning Fundamentals, Part I+

   Large Language Models (LLMs), supervised and unsupervised learning, evaluation metrics, and model generalization.

2. L2Deep Learning Fundamentals, Part II+

   Perceptrons, multilayer perceptrons (MLPs), gradient descent, backpropagation, and loss minimization.

3. L3Data Preparation for LLMs+

   Embeddings, Word2Vec, and the step-by-step process for preparing embeddings.

4. L4Attention+

   Recurrent neural networks, the information bottleneck, the attention mechanism, and their variants.

5. L5Transformer Architecture+

   Transformers vs. RNNs, token and positional embeddings, the Transformer block, layer normalization, feed-forward and GELU layers, residual connections, the GPT-2 architecture, text generation, and encoder, decoder, and encoder-decoder models.

6. L6Training+

   Pretraining with BERT and GPT, masked and next-token prediction, contextual embeddings, sampling techniques, fine-tuning for classification, adapters and LoRA, instruction fine-tuning, and RLHF.

7. L7Prompting+

   Prompt engineering, temperature and max tokens, zero-shot and few-shot prompting, chain-of-thought, self-correction and reflection, context windows, KV cache, prompt caching, positional bias, hallucinations, adversarial prompting, biases and security risks.

8. L8APIs and model routing+

   REST APIs, methods and status codes, JSON responses, Postman, Python Requests, OpenAI and Anthropic SDKs, streaming, the routing trilemma, and task-based, complexity-based, fallback, and semantic routing.

9. L9Reasoning models and modern architectures+

   Supervised fine-tuning, reward modeling, RLHF with PPO, DPO, RL for reasoning, latent reasoning tokens, process and outcome rewards, test-time compute, DeepSeek R1, Mixture of Experts, open-weight models, licensing, and data sovereignty.

• Advanced RAG
• GraphRAG
• Hybrid search
• Agent orchestration
• MCP integration
• RAG evaluation

1. L1Structured and unstructured data for LLMs+

   Data for LLMs, data warehousing, relational schemas, SQL queries and joins, MongoDB documents, Text-to-SQL agents, and SQL guardrails.

2. L2Vector databases and the RAG pipeline+

   Information retrieval, BM25, the retrieval pipeline, chunking strategies, embeddings, similarity metrics, vector databases, retrieval-augmented generation, and precision and recall at k.

3. L3Knowledge Graphs and GraphRAG+

   Multi-hop queries, entities and relations, RDF vs. property graphs, entity extraction with LLMs, entity resolution, Neo4j and Cypher, Text-to-Cypher, and global vs. local GraphRAG.

4. L4Advanced Retrieval+

   The quality gap in retrieval, hybrid search, score normalization, reciprocal rank fusion, cross-encoders, ColBERT and late interaction, LLM reranking, embedding fine-tuning, and hard-negative mining.

5. L5Monitoring and evaluation of RAG systems+

   Retrieval metrics, nDCG and MRR, the RAG triad, Ragas metrics, synthetic test sets, LLM-as-a-judge, judge calibration, monitoring vs. observability, LangSmith, Langfuse, and DeepEval, and token cost tracking.

6. L6Agents, tools, and MCP+

   Agents vs. workflows, ReAct, function calling, tool design, structured outputs, the agent loop, agentic RAG, the Model Context Protocol, and MCP security.

7. L7Agentic loops and execution environments+

   State, nodes, and edges in LangGraph, tool binding, ToolNode, prebuilt agents, ReAct, routers, parallelization, LLM-as-judge, reflection, human-in-the-loop, sandboxes, the tool invocation runtime, and security in code execution.

8. L8The agent harness+

   Context degradation, episodic, semantic, and procedural memory, Claude Code Skills, short and long-term memory, compaction, checkpointers, tracing, and tokenomics.

9. L9Deep Agents and SDKs+

   Planning with to-dos, sub-agents, virtual file systems, Open Deep Researcher, the Deep Agents library, the Claude Agent SDK, and how to choose an agent framework.

• LLMOps
• CI/CD
• Cloud deployment
• Evals
• Guardrails
• Production security

1. L1From Agents to Services with FastAPI+

   MLOps vs. LLMOps, HTTP, REST APIs, status codes, agents with LangGraph, FastAPI, validation with Pydantic, health checks, streaming, ASGI and Uvicorn, and configuration.

2. L2Introduction to Docker+

   Containers vs. virtual machines, namespaces, cgroups, Dockerfiles, base images, .dockerignore, layers, build cache, multi-stage builds, ports, volumes, registries, tags, and digests.

3. L3Docker Compose and the multi-service stack+

   Compose files, services, networks, ports vs. expose, named volumes, bind mounts, backups, healthchecks, startup order, debugging, response caching with Redis, override files, and .env files.

4. L4EC2 and server administration for LLMOps+

   AWS accounts, IAM, MFA, budgets, billing, EC2 instances, instance types, Elastic IPs, security groups, SSH, Linux administration, permissions, Docker on EC2, crash recovery, Caddy, HTTPS, and DNS.

5. L5Amazon Bedrock and a vendor-agnostic architecture+

   Amazon Bedrock, model IDs, inference profiles, IAM roles, provider abstraction, get_llm(), sampling parameters, retries, fallback, price per token, cost estimation, model selection, streaming, and managed vs. self-hosted inference.

6. L6CI/CD for Agentic Systems+

   Prompt and model versioning, CI vs. CD, pull requests, GitHub Actions, runners, secrets, pipeline caching, text output testing, unit tests vs. evals, eval gates, immutable tags, SSH deployment, rollback, staging, canary, and blue-green.

7. L7User interfaces for agents+

   Streamlit, Gradio, session state, streaming responses, error states, visible reasoning, tool call rendering, human-in-the-loop, LangGraph interrupts, checkpointers, designing for trust, access control, and UI deployment.

8. L8Observability and the feedback loop+

   Logs, traces, metrics, structured logging, log levels, correlation IDs, Langfuse, sampling, retention, dashboards, alerts, SLOs, cost attribution, continuous evaluation, LLM-as-a-judge, feedback capture, and feedback ethics.

9. L9Guardrails and security in production+

   Prompt injection, threat modeling, input validation, output validation, guard models, tool design, Bedrock Guardrails, NeMo Guardrails, secrets management, least privilege, container hardening, API authentication, rate limiting, audit logs, and incident response.

Team-based industry project: from a partner company or assigned by Anyone AI.

It closes the Fellowship, right after the final project ends.