Elliot O
Roadmaps

AI Engineering Roadmap

A practical path through foundations, context, retrieval, agents, evaluation, and operations. Built from The Modern Engineer archive.

6 stages48 issues

Stage 01

Foundation and System Judgment

7 issues
0%

Understand what AI engineering owns: behavior, contracts, limits, and proof.

Outcome

You can describe an AI system as components, controls, and failure modes.

AI engineering scopemodel tradeoffsstructured outputsprompt formatsportfolio proof
Developer WorkflowAI-Assisted Coding and Emerging Software Engineering TrendsAI FoundationsAI Engineering Roadmap 2026AI FoundationsClassical Machine Learning vs Modern AI SystemsAI FoundationsStructured Inputs and Outputs for Deterministic LLM SystemsAI FoundationsPrompt Format Languages for AI EngineeringDeveloper WorkflowThe Portfolio That Proves You Can Build AI SystemsEvaluation & GovernanceChain-of-Thought Control Contracts for AI Systems

Stage 02

Prompt, Context and Memory

7 issues
0%

Treat prompts, context, and memory as engineered inputs.

Outcome

You can shape model behavior without hiding system policy inside loose prose.

prompt designcontext budgetsmemory boundarieslineageagent-readable docs
AI FoundationsPrompt Engineering FundamentalsRetrieval & ContextContext Engineering FundamentalsRetrieval & ContextDeterministic Context Budgeting for Reliable LLM PromptingRetrieval & ContextProduction Memory Architecture for AI SystemsRetrieval & ContextSession Memory and Context Compaction for Production AI AgentsEvaluation & GovernancePrompt and Context Lineage for Reproducible AI SystemsDeveloper WorkflowAgent-Readable Documentation and Context Packs

Stage 03

Retrieval and Knowledge Systems

7 issues
0%

Design systems that retrieve useful evidence and reject weak context.

Outcome

You can build RAG and vector-search flows with clear data contracts.

RAGvector searchsemantic searchdata contractsindex lifecycle
Retrieval & ContextLocal RAG System using Semantic Kernel, Ollama, and QdrantRetrieval & ContextCategory-Aware Local RAG System using ASP.NET Core MVC, Ollama, and pgvectorRetrieval & ContextBuilding a Local Semantic Runbook Search Engine with Ollama and Microsoft.Extensions.AIRetrieval & ContextDeterministic Semantic Retrieval with Embeddings and Vector SearchRetrieval & ContextAI Data Contracts for RAG and Agent ContextRetrieval & ContextPoint-in-Time Feature Stores for Hybrid AI DecisionsRetrieval & ContextEmbedding Index Lifecycle Gates for RAG Systems

Stage 04

Agents and Workflow Control

10 issues
0%

Move from single model calls to bounded workflows with tools, state, and control flow.

Outcome

You can define agent loops, tool boundaries, orchestration, and typed decisions.

Semantic KernelMCPintent routingdurable agentstyped decisions
Agents & WorkflowsHybrid AI Workflows with Semantic KernelAgents & WorkflowsBuilding a Modern MCP Server and Client with Ollama for Deterministic AI WorkflowsAgents & WorkflowsBuilding a Reactive Local AI Agent with Semantic Kernel and OllamaAgents & WorkflowsIntent Routing and Deterministic AI Control FlowAgents & WorkflowsProduction-Grade AI Agents with Microsoft Agent Framework and Deterministic GuardrailsAgents & WorkflowsGrounded Multi-Agent Research Review with Microsoft Agent FrameworkAgents & WorkflowsDurable Agent Workflows with Policy-Bounded ExecutionAgents & WorkflowsAgent Interoperability with A2A and Microsoft Foundry on Azure Container AppsAgents & WorkflowsDesigning Control Loops for AI AgentsAgents & WorkflowsJev: A New Kind of AI Model for Typed Decisions

Stage 05

Evaluation, Safety and Governance

11 issues
0%

Make behavior measurable before trusting it in production.

Outcome

You can add evaluation gates, observability, approval points, and release checks.

guardrailsevaluationmodel gatestool contractsruntime governance
Evaluation & GovernanceGuardrails, Constraints, and Failure Modes in Production AI SystemsEvaluation & GovernanceEvaluation, Observability, and Feedback Loops in Production AI SystemsEvaluation & GovernancePrompt Versioning and A/B EvaluationEvaluation & GovernanceTrusted Agent Operations with Least-Privilege Tools and PII-Safe ContextEvaluation & GovernanceModel Upgrade Gates for AI SystemsEvaluation & GovernanceMCP Tool Contract Gates for AI SystemsEvaluation & GovernanceHuman Approval Gates for High-Risk AI ActionsEvaluation & GovernanceTwo-Layer Content Moderation Gates for AI SystemsEvaluation & GovernanceGolden Dataset Governance for AI Evaluation SystemsEvaluation & GovernanceTool Result Admission Gates for Production AI AgentsEvaluation & GovernanceFeature-Flagged AI Runtime Controls

Stage 06

Runtime and Engineering Operations

6 issues
0%

Run AI systems with reliability, cost awareness, and delivery discipline.

Outcome

You can operate AI workflows like production software, not demos.

API reliabilitycost controlstreamingbatch jobsrepository control
Developer WorkflowSpec-Driven AI-Assisted Coding and Agent ControlRuntime & OperationsProduction Reliability Engineering for AI APIsRuntime & OperationsAI Cost Engineering for Local InferenceDeveloper WorkflowAGENTS.md and the Repository Control Plane for Coding AgentsRuntime & OperationsStreaming AI GatewaysRuntime & OperationsBatch Inference Job Control for AI Pipelines