AI ENGINEER ROADMAP / A LEARNING JOURNEY

From Software Engineer
to AI Engineer.

Your journey from writing traditional software to engineering intelligent, production-ready AI systems. Level up your skills, complete missions, and build your way to becoming an AI Engineer.

9 LEVELS · 65 MISSIONS (9 PUBLISHED · 56 IN WRITING) · 100 XP PER MISSION · 900 XP TOTAL

YOUR PROGRESS

0 of 0 missions

0 / 0 XP

START YOUR JOURNEY

NO PREREQUISITES — OPEN ANY PUBLISHED MISSION IN ANY ORDER

START HERE · The Software Engineer Foundation

Understand how existing software engineering skills transfer to AI engineering.

4 TOPICS · 1 PUBLISHED

BADGE Foundation

0 / 1 MISSIONS

MISSION 0.1 PUBLISHED What Is an AI Engineer? Roles, Responsibilities, and Career PathsWhat the role actually is beyond the job-title noise — what AI engineers build all day, how the role differs from ML engineering and data science, and the career ladder from junior to principal.LVL 0 · FoundationsOPEN MISSION →

MISSION 0.2COMING SOONSoftware Engineer vs. AI Engineer: What Actually Changes?A honest diff of the two roles: what transfers directly, what gets re-weighted, and which habits from traditional software engineering become superpowers in AI work.LVL 0 · FoundationsIN WRITING

MISSION 0.3COMING SOONThe AI Engineer Skill MapThe full competency map — Python, data, ML fundamentals, LLM engineering, RAG, agents, evaluation, and cloud — organized by depth so you can plan what to learn next.LVL 0 · FoundationsIN WRITING

MISSION 0.4COMING SOONSetting Up Your AI Engineering EnvironmentA pragmatic local setup: Python version management, virtual environments, API keys done safely, model runners, and the editor tooling that keeps AI experiments reproducible.LVL 0 · FoundationsIN WRITING

Python & Data Foundations

Develop the programming and data skills needed for AI development.

6 TOPICS · 1 PUBLISHED

BADGE Python & Data

0 / 1 MISSIONS

MISSION 1.1 PUBLISHED Python for Experienced Software EngineersPython for developers who already write C#, Java, or TypeScript — tooling, project layout, typing, testing, and the idioms worth adopting (and the ones to skip).LVL 1 · Python & DataOPEN MISSION →

MISSION 1.2COMING SOONPython Type Hints, Pydantic, and Data ValidationTreat untyped Python like legacy code: type hints, mypy, and Pydantic models as the contract layer that keeps LLM inputs and outputs honest.LVL 1 · Python & DataIN WRITING

MISSION 1.3COMING SOONAsync Programming, APIs, and Concurrencyasyncio without the folklore — when async actually helps AI workloads, batching LLM calls, timeouts, retries, and backpressure for API-bound systems.LVL 1 · Python & DataIN WRITING

MISSION 1.4COMING SOONNumPy, Pandas, and Data ProcessingVectorized thinking for application engineers: arrays, dataframes, joins, and the data-wrangling patterns that feed embeddings, evaluation sets, and fine-tuning data.LVL 1 · Python & DataIN WRITING

MISSION 1.5COMING SOONEssential Mathematics and Statistics for AIJust enough math to be dangerous: vectors and dot products, probability distributions, sampling, and the statistics behind evaluation metrics — with engineer-friendly intuition.LVL 1 · Python & DataIN WRITING

MISSION 1.6COMING SOONPython Testing, Packaging, and Project Structurepytest, fixtures, and property-based tests for non-deterministic systems; packaging with pyproject.toml; and a project structure that survives contact with a real team.LVL 1 · Python & DataIN WRITING

Machine Learning Fundamentals

Understand how machine learning models learn and make predictions.

6 TOPICS · 1 PUBLISHED

BADGE ML Fundamentals

0 / 1 MISSIONS

MISSION 2.1 PUBLISHED AI vs. Machine Learning vs. Deep Learning vs. Generative AIThe nested-sets mental model that ends the terminology confusion — what each term actually means, where LLMs sit, and how to choose between predictive ML and generative approaches.LVL 2 · ML BasicsOPEN MISSION →

MISSION 2.2COMING SOONSupervised, Unsupervised, and Reinforcement LearningThe three learning paradigms explained through engineering problems you already know — classification, clustering, and reward-driven behavior — and where each shows up in AI products.LVL 2 · ML BasicsIN WRITING

MISSION 2.3COMING SOONTraining, Validation, and Test DatasetsWhy splitting data correctly is the difference between a model that works and one that merely memorized — leakage, stratification, and the discipline of a untouched test set.LVL 2 · ML BasicsIN WRITING

MISSION 2.4COMING SOONOverfitting, Underfitting, and GeneralizationThe core failure mode of every learned system: memorizing the past versus generalizing to the future — diagnostics, regularization, and what it means for fine-tuning LLMs.LVL 2 · ML BasicsIN WRITING

MISSION 2.5COMING SOONMachine Learning Evaluation MetricsAccuracy is a trap. Precision, recall, F1, ROC-AUC, and calibration — which metric matches which business failure, and how to read them like an engineer reads a benchmark.LVL 2 · ML BasicsIN WRITING

MISSION 2.6COMING SOONBuild Your First Machine Learning ProjectAn end-to-end guided project: frame the problem, prepare data, train a baseline, evaluate honestly, and ship a prediction endpoint — the full loop in one sitting.LVL 2 · ML BasicsIN WRITING

Deep Learning & Transformers

Understand the foundations of modern language models.

8 TOPICS · 2 PUBLISHED

BADGE Transformers

0 / 2 MISSIONS

MISSION 3.1COMING SOONNeural Networks Explained for Software EngineersNeural networks as trainable functions: layers, weights, activation, loss, and gradient descent — explained with the mental models of a developer, not a mathematician.LVL 3 · Deep LearningIN WRITING

MISSION 3.2 PUBLISHED Embeddings: Turning Meaning into VectorsHow text becomes coordinates in a space where distance means similarity — the foundation of semantic search, RAG, clustering, and every modern retrieval system.LVL 3 · Deep LearningOPEN MISSION →

MISSION 3.3COMING SOONAttention Mechanisms and TransformersThe architecture behind the LLM era: self-attention, multi-head attention, and why transformers replaced recurrence — explained structurally, not symbolically.LVL 3 · Deep LearningIN WRITING

MISSION 3.4COMING SOONTokens, Tokenization, and Context WindowsWhy models see text as tokens, how tokenizers work, and what context windows really cost — the mechanics behind token limits, pricing, and prompt budgeting.LVL 3 · Deep LearningIN WRITING

MISSION 3.5COMING SOONHow Large Language Models WorkPretraining, next-token prediction, and instruction tuning — a working mental model of what the model is actually doing when it answers, and why it hallucinates.LVL 3 · Deep LearningIN WRITING

MISSION 3.6COMING SOONModel Parameters, Weights, and InferenceWhat "7B parameters" means physically, how weights are stored and loaded, and what happens during inference — the hardware reality behind every API call.LVL 3 · Deep LearningIN WRITING

MISSION 3.7 PUBLISHED Quantization: FP32, FP16, INT8, and 4-Bit ModelsStoring model weights with fewer bits: the memory math, the quality trade-offs, and how 4-bit quantization makes large models run on ordinary hardware.LVL 3 · Deep LearningOPEN MISSION →

MISSION 3.8COMING SOONLocal Models, GPU Memory, and Model ServingRunning models on your own hardware: VRAM budgeting, inference servers, batching, and when local models beat hosted APIs — and when they absolutely do not.LVL 3 · Deep LearningIN WRITING

LLM Application Engineering

Build practical applications powered by large language models.

7 TOPICS · 0 PUBLISHED

BADGE LLM Engineering

0 / 0 MISSIONS

MISSION 4.1COMING SOONLLM APIs: Prompts, Messages, and Model ParametersThe programming interface of LLMs: message roles, system prompts, temperature, top-p, max tokens, and stop sequences — the API surface every AI engineer must know cold.LVL 4 · LLM AppsIN WRITING

MISSION 4.2COMING SOONPrompt Engineering Beyond Basic PromptingStructured prompting as interface design: role framing, few-shot examples, output contracts, and decomposition — treated like API design, not incantation.LVL 4 · LLM AppsIN WRITING

MISSION 4.3COMING SOONStructured Outputs, JSON, and Function CallingGetting reliable data out of probabilistic models: JSON mode, schema-constrained outputs, and function calling — the difference between a demo and an integration.LVL 4 · LLM AppsIN WRITING

MISSION 4.4COMING SOONBuilding LLM Applications with Python and FastAPIA production-shaped LLM service: FastAPI endpoints, streaming responses, configuration, secrets, and the request lifecycle of an AI-backed API.LVL 4 · LLM AppsIN WRITING

MISSION 4.5COMING SOONStreaming Responses and Conversation ManagementServer-sent events, token streaming, conversation state, truncation strategies, and multi-turn memory — making chat feel instant without losing the thread.LVL 4 · LLM AppsIN WRITING

MISSION 4.6COMING SOONHosted Models vs. Open-Weight ModelsBedrock, Azure OpenAI, and Anthropic versus Llama, Mistral, and Qwen: cost, latency, privacy, control, and the decision framework for hosted versus self-hosted.LVL 4 · LLM AppsIN WRITING

MISSION 4.7COMING SOONManaging Latency, Cost, and ReliabilityThe production triangle of LLM applications: token-level latency budgets, cost per request, retries, fallbacks, and graceful degradation when the model misbehaves.LVL 4 · LLM AppsIN WRITING

RAG & Knowledge-Based AI

Build AI applications that retrieve information from enterprise knowledge sources.

8 TOPICS · 1 PUBLISHED

BADGE RAG Engineer

0 / 1 MISSIONS

MISSION 5.1 PUBLISHED What Is RAG, and Why Do AI Applications Need It?What RAG is, how it actually works in production, the trade-offs most teams miss, and how retrieval quality is approached in regulated enterprise systems.LVL 5 · RAGOPEN MISSION →

MISSION 5.2COMING SOONDocument Ingestion, Parsing, and ChunkingThe unglamorous foundation of every good RAG system: parsing PDFs and HTML, structure-aware chunking, overlap strategies, and metadata extraction that survives audits.LVL 5 · RAGIN WRITING

MISSION 5.3COMING SOONEmbeddings and Vector Similarity SearchEmbedding models for retrieval: choosing dimensions, comparing similarity metrics, normalizing vectors, and measuring whether your search actually finds the right things.LVL 5 · RAGIN WRITING

MISSION 5.4COMING SOONVector Databases: Milvus, pgvector, and AlternativesChoosing and operating a vector store: pgvector for Postgres shops, Milvus at scale, indexes (HNSW, IVF), filtering, and the operational realities nobody blogs about.LVL 5 · RAGIN WRITING

MISSION 5.5COMING SOONBuilding an End-to-End RAG PipelineIngestion to answer: a complete reference pipeline with chunking, embedding, hybrid retrieval, reranking, context construction, and cited generation.LVL 5 · RAGIN WRITING

MISSION 5.6COMING SOONHybrid Search, Metadata Filtering, and RerankingWhere retrieval quality is actually won: combining vector and keyword search, enforcing permissions with metadata filters, and rerankers that pay for themselves.LVL 5 · RAGIN WRITING

MISSION 5.7COMING SOONRAG Evaluation: Faithfulness, Relevance, Precision, and RecallMeasuring retrieval and generation separately: hit rate, MRR, faithfulness, and answer correctness — with golden sets and regression gates that make quality a number.LVL 5 · RAGIN WRITING

MISSION 5.8COMING SOONAdvanced RAG: Query Rewriting and Retrieval OptimizationBeyond naive top-k: query expansion, HyDE, multi-query retrieval, parent-document chunking, and the optimization patterns that fix real retrieval failures.LVL 5 · RAGIN WRITING

Agentic AI & Workflow Orchestration

Build controlled, multi-step AI workflows and tool-using agents.

8 TOPICS · 2 PUBLISHED

BADGE Agent Architect

0 / 2 MISSIONS

MISSION 6.1 PUBLISHED Agentic AI: Agents vs. WorkflowsWhat agentic AI actually is beyond the hype — stateful workflows, tools, HITL gates — and how to design agent systems that survive production in regulated domains.LVL 6 · AgentsOPEN MISSION →

MISSION 6.2COMING SOONTool Calling and AI AgentsTools are the agent’s hands: designing schema-validated tool contracts, handling call failures, and why tool quality — not model quality — determines agent reliability.LVL 6 · AgentsIN WRITING

MISSION 6.3COMING SOONState, Memory, and PlanningThe stateful core of agentic systems: typed task state, short-term versus long-term memory, and planning strategies that stay inspectable and debuggable.LVL 6 · AgentsIN WRITING

MISSION 6.4COMING SOONLangGraph: Nodes, Edges, State, and CheckpointingOrdinary engineering concepts applied to LLM applications: graphs of typed steps, conditional edges, persisted checkpoints, and resume-able long-running workflows.LVL 6 · AgentsIN WRITING

MISSION 6.5COMING SOONHuman-in-the-Loop WorkflowsThe single most effective trust mechanism in enterprise AI: approval gates, interrupt/resume patterns, and designing review surfaces experts actually accept.LVL 6 · AgentsIN WRITING

MISSION 6.6COMING SOONMulti-Agent Systems: When and Why to Use ThemSplitting work across agents only when responsibilities genuinely differ — generator versus critic patterns, coordination costs, and the multi-agent decision test.LVL 6 · AgentsIN WRITING

MISSION 6.7 PUBLISHED MCP: Model Context Protocol ExplainedWhy MCP matters as enterprise AI infrastructure — standardized tool surfaces, retrieval as a contract — and how an MCP-based semantic retrieval platform is architected.LVL 6 · AgentsOPEN MISSION →

MISSION 6.8COMING SOONBuild a Production-Style AI AgentA guided capstone: a LangGraph agent with typed state, validated tools, HITL gates, tracing, and evaluation — the shape of a real enterprise agent, not a toy loop.LVL 6 · AgentsIN WRITING

Production AI Engineering

Build secure, reliable, observable, and scalable AI systems.

9 TOPICS · 1 PUBLISHED

BADGE Production AI

0 / 1 MISSIONS

MISSION 7.1 PUBLISHED LLM Evaluation: Golden Datasets and Automated TestingHow to evaluate LLM systems like an engineer — golden sets, per-node metrics, regression gates — and why vibes-based quality is the most expensive mistake in AI engineering.LVL 7 · ProductionOPEN MISSION →

MISSION 7.2COMING SOONGuardrails, Hallucination Mitigation, and Output ValidationDefense in depth for generated content: grounding checks, schema validation, refusal design, and guardrail layers that catch failures before users do.LVL 7 · ProductionIN WRITING

MISSION 7.3COMING SOONAI Observability with LangSmith and CloudWatchTracing every node, tool call, and token: LangSmith for LLM-native traces, CloudWatch for the platform around them, and the dashboards that make AI systems debuggable.LVL 7 · ProductionIN WRITING

MISSION 7.4COMING SOONPrompt Injection and AI Application SecurityThe new attack surface: direct and indirect prompt injection, tool abuse, data exfiltration through context, and the structural defenses that actually hold.LVL 7 · ProductionIN WRITING

MISSION 7.5COMING SOONAuthentication, Authorization, Secrets, and Data PrivacyAI applications are still applications: scoped credentials per tool, retrieval as a data-access surface, PII lifecycle in vector indexes, and privacy by architecture.LVL 7 · ProductionIN WRITING

MISSION 7.6COMING SOONCost Optimization, Caching, and Rate LimitingToken economics as an engineering discipline: semantic caches, prompt caching, model tiering, rate limits, and the cost model that keeps AI features viable.LVL 7 · ProductionIN WRITING

MISSION 7.7COMING SOONDeployment with Docker, AWS, and CI/CDShipping AI services like software: containerized FastAPI services, ECS deployment, prompt and evaluation gates in CI, and infrastructure as code for AI workloads.LVL 7 · ProductionIN WRITING

MISSION 7.8COMING SOONMonitoring, Tracing, and Production DebuggingWhen quality drops at 2 a.m.: trace-first debugging, retrieval logging, drift detection, and the incident playbook for systems that fail in novel ways.LVL 7 · ProductionIN WRITING

MISSION 7.9COMING SOONResponsible AI, Governance, and Human OversightBuilding AI systems organizations can defend: model documentation, audit trails, human oversight, and governance that satisfies regulators without killing velocity.LVL 7 · ProductionIN WRITING

FINAL MILESTONE · Advanced AI Systems & Career Readiness

Develop the skills and engineering judgment expected of senior AI engineers.

9 TOPICS · 0 PUBLISHED

BADGE Senior AI Engineer

0 / 0 MISSIONS

MISSION 8.1COMING SOONAI System Design: From Requirements to ArchitectureA design method for AI systems: translating business requirements into architecture decisions — model choice, retrieval, orchestration, and failure strategy.LVL 8 · MasteryIN WRITING

MISSION 8.2COMING SOONDesigning Scalable RAG and Agentic SystemsScaling the patterns from earlier levels: multi-tenant retrieval, tiered indexes, async orchestration, and the architecture of AI platforms serving thousands of users.LVL 8 · MasteryIN WRITING

MISSION 8.3COMING SOONModel Selection, Fine-Tuning, and When Not to Fine-TuneThe decision framework senior engineers actually use: prompting versus RAG versus fine-tuning, evaluation-driven selection, and the fine-tuning trap most teams fall into.LVL 8 · MasteryIN WRITING

MISSION 8.4COMING SOONLoRA, QLoRA, and Parameter-Efficient Fine-TuningAdapting models without retraining them: low-rank adapters, quantized LoRA, and PEFT — how small matrices change big models, and when the effort is worth it.LVL 8 · MasteryIN WRITING

MISSION 8.5COMING SOONQuantized Models and Inference OptimizationProduction inference engineering: quantization strategies, KV-cache management, speculative decoding, and squeezing latency and cost out of every token.LVL 8 · MasteryIN WRITING

MISSION 8.6COMING SOONAI Workload Deployment on AWSBedrock, SageMaker, ECS, and Lambda for AI: reference architectures for RAG platforms, agent services, and batch inference on AWS — with cost and security in mind.LVL 8 · MasteryIN WRITING

MISSION 8.7COMING SOONBuild an End-to-End Enterprise AI CapstoneThe final mission: design, build, evaluate, and document a complete enterprise AI system — RAG plus agents plus evaluation plus deployment — as a portfolio centerpiece.LVL 8 · MasteryIN WRITING

MISSION 8.8COMING SOONAI Engineering Interviews: System Design and Technical DiscussionsHow AI engineering interviews actually run: system design prompts, trade-off discussions, evaluation questions, and answering like an engineer instead of a tutorial.LVL 8 · MasteryIN WRITING

MISSION 8.9COMING SOONBuilding a Portfolio That Demonstrates Senior-Level AI EngineeringTurning projects into evidence: case studies that show architecture and trade-offs, evaluation results, and the portfolio signals that separate senior from junior.LVL 8 · MasteryIN WRITING