How to use it
Work through the roadmap in order. Parts I to III build one system — a document-grounded health assistant with an agent and an MCP server in front of it — and Parts IV and V add the machine learning and production work around it.
Click the box beside a topic to move it from not started to in progress to done. Every topic also takes a note: what you built, the link, or what you still do not understand.
Not startedIn progressDone
AI Engineering Foundation
0 / 740/8
- 1.1AI Service Architecture
- 1.2REST API Design
- 1.3Dependency Injection
- 1.4Database Schema Design
- 1.5Transactions and Connection Pooling
- 1.6Flyway Migrations
- 1.7JSONB and Structured Data
- 1.8Error Handling
0/9
- 2.1PDF Parsing
- 2.2Text Extraction
- 2.3OCR Fundamentals
- 2.4Digital vs. Scanned PDFs
- 2.5Table Extraction
- 2.6Document Metadata
- 2.7Document Validation
- 2.8Malformed Document Handling
- 2.9Extraction Quality
0/10
- 3.1Embedding Fundamentals
- 3.2Embedding Models
- 3.3Vector Dimensions
- 3.4Cosine Similarity
- 3.5Distance Metrics
- 3.6Vector Storage
- 3.7pgvector Operators
- 3.8HNSW Indexing
- 3.9IVFFlat Indexing
- 3.10Index Performance Trade-offs
0/9
- 4.1Document Chunking
- 4.2Chunk Size and Overlap
- 4.3Metadata Filtering
- 4.4Top-K Retrieval
- 4.5Similarity Thresholds
- 4.6Hybrid Search
- 4.7Keyword vs. Vector Search
- 4.8Reranking
- 4.9Retrieval Optimization
0/10
- 5.1RAG Architecture
- 5.2Document Ingestion Pipeline
- 5.3Query Processing
- 5.4Retrieval Pipeline
- 5.5Context Assembly
- 5.6Prompt Construction
- 5.7Source Attribution
- 5.8Hallucination Prevention
- 5.9Context Window Management
- 5.10No-Answer Handling
0/9
- 6.1Golden Datasets
- 6.2Retrieval Precision
- 6.3Retrieval Recall
- 6.4Hit Rate
- 6.5Mean Reciprocal Rank (MRR)
- 6.6Answer Correctness
- 6.7Faithfulness
- 6.8Groundedness
- 6.9Regression Testing
0/10
- 7.1Spring Boot Actuator
- 7.2Micrometer
- 7.3Structured Logging
- 7.4Distributed Tracing
- 7.5Latency Metrics
- 7.6Token Usage
- 7.7Cost Tracking
- 7.8Retrieval Metrics
- 7.9Error Monitoring
- 7.10Correlation IDs
0/9
- 8.1Structured Health Events
- 8.2Event Extraction
- 8.3Temporal Data Modeling
- 8.4Observed Date vs. Ingestion Date
- 8.5Source Provenance
- 8.6Timeline APIs
- 8.7Temporal Retrieval
- 8.8Conflicting Health Values
- 8.9Historical Context Management
Agent Engineering
0 / 840/5
- 9.1LLM Applications vs. AI Agents
- 9.2Agent Architecture
- 9.3Observe–Decide–Act Cycle
- 9.4Deterministic vs. Agentic Workflows
- 9.5Agent Design Patterns
0/10
- 10.1Tool Fundamentals
- 10.2Tool vs. API vs. Function
- 10.3Tool Contracts
- 10.4Input Schemas
- 10.5Output Schemas
- 10.6Tool Descriptions
- 10.7Tool Registry
- 10.8Read-Only vs. State-Changing Tools
- 10.9Tool Errors and Timeouts
- 10.10Idempotency and Retries
0/11
- 11.1Model-Initiated Tool Calls
- 11.2Function Calling vs. JSON Generation
- 11.3JSON Schema
- 11.4Tool-Calling Request/Response Flow
- 11.5Tool-Call IDs
- 11.6Argument Validation
- 11.7Tool Execution in Java
- 11.8Returning Tool Results
- 11.9Parallel Tool Calls
- 11.10Invalid Tool Arguments
- 11.11Structured Outputs vs. Tool Calling
0/12
- 12.1Agent Execution Cycle
- 12.2Model Response Types
- 12.3State Transitions
- 12.4Loop Termination
- 12.5Maximum Iterations
- 12.6Execution Deadlines
- 12.7Token and Cost Budgets
- 12.8Repeated-Call Detection
- 12.9No-Progress Detection
- 12.10Partial Results
- 12.11Graceful Termination
- 12.12Synchronous vs. Asynchronous Execution
0/15
- 13.1Conversation History
- 13.2Agent Execution State
- 13.3Short-Term Memory
- 13.4Long-Term Memory
- 13.5Working State and Tool Results
- 13.6Persistent State in PostgreSQL
- 13.7Checkpointing
- 13.8State Recovery
- 13.9Thread and Conversation IDs
- 13.10Memory Summarization
- 13.11Relevant-Memory Retrieval
- 13.12Temporal Memory
- 13.13Conflicting and Outdated Facts
- 13.14Memory Isolation
- 13.15State Expiration and Retention
0/15
- 14.1Task-Success Rate
- 14.2Tool-Selection Accuracy
- 14.3Tool-Argument Correctness
- 14.4Tool Execution Success
- 14.5Multi-Step Completion Rate
- 14.6Final-Answer Correctness
- 14.7Groundedness and Citation Accuracy
- 14.8Agent Efficiency
- 14.9Latency and Cost Evaluation
- 14.10Deterministic Testing
- 14.11LLM-as-Judge
- 14.12Golden Test Scenarios
- 14.13Adversarial Evaluation
- 14.14Failure Classification
- 14.15Regression Testing
0/16
- 15.1AI Threat Modeling
- 15.2Direct Prompt Injection
- 15.3Indirect Prompt Injection
- 15.4Untrusted Document Content
- 15.5Tool Misuse
- 15.6Privilege Escalation
- 15.7Least-Privilege Design
- 15.8Input and Output Validation
- 15.9Sensitive Data Leakage
- 15.10Cross-User Data Isolation
- 15.11Secret Management
- 15.12PII Protection
- 15.13Unsafe Operations
- 15.14Approval Gates
- 15.15Denial-of-Service Protection
- 15.16Security Testing and Audit Trails
Model Context Protocol (MCP)
0 / 1080/12
- 16.1What Is MCP?
- 16.2Problems MCP Solves
- 16.3MCP vs. REST APIs
- 16.4MCP vs. Function Calling
- 16.5MCP Host
- 16.6MCP Client
- 16.7MCP Server
- 16.8JSON-RPC 2.0
- 16.9Requests, Responses and Notifications
- 16.10Protocol Initialization
- 16.11Capability Negotiation
- 16.12Protocol Versioning
0/10
- 17.1Host–Client–Server Architecture
- 17.2Communication Lifecycle
- 17.3Capability Discovery
- 17.4Message Framing
- 17.5Standard I/O (stdio)
- 17.6Streamable HTTP
- 17.7Local vs. Remote Servers
- 17.8Session Management
- 17.9Multiple MCP Servers
- 17.10Reconnection and Failure Handling
0/13
- 18.1MCP SDK Fundamentals
- 18.2Java MCP SDK
- 18.3Server Initialization
- 18.4Server Metadata
- 18.5Capability Declaration
- 18.6Request Routing
- 18.7Tool Registration
- 18.8Resource Registration
- 18.9Prompt Registration
- 18.10Error Handling
- 18.11Logging and Debugging
- 18.12Graceful Shutdown
- 18.13Protocol Integration Testing
0/13
- 19.1Tool Discovery
- 19.2tools/list
- 19.3tools/call
- 19.4Tool Names and Descriptions
- 19.5Input Schemas
- 19.6Argument Validation
- 19.7Structured Tool Results
- 19.8Error Results
- 19.9Tool Annotations
- 19.10Authorization
- 19.11Long-Running Operations
- 19.12Timeouts and Retries
- 19.13Tool Versioning
0/13
- 20.1Resource Fundamentals
- 20.2Resources vs. Tools
- 20.3Resource URIs
- 20.4resources/list
- 20.5resources/read
- 20.6Static Resources
- 20.7Dynamic Resources
- 20.8Resource Templates
- 20.9MIME Types and Metadata
- 20.10Resource Subscriptions
- 20.11Change Notifications
- 20.12Pagination and Large Resources
- 20.13Resource Access Control
0/11
- 21.1Prompt Fundamentals
- 21.2Prompts vs. System Instructions
- 21.3prompts/list
- 21.4prompts/get
- 21.5Prompt Metadata
- 21.6Prompt Arguments
- 21.7Prompt Templates
- 21.8Message Roles
- 21.9Dynamic Prompt Construction
- 21.10Prompt Versioning
- 21.11Prompt Testing
0/12
- 22.1Defining Server Boundaries
- 22.2Reusing Spring Services
- 22.3MCP SDK Integration
- 22.4Transport Configuration
- 22.5Health Tool Registration
- 22.6Resource Implementation
- 22.7Prompt Implementation
- 22.8Service and Protocol Error Mapping
- 22.9Authentication and Authorization
- 22.10Logging and Tracing
- 22.11Integration Testing
- 22.12Packaging and Deployment
0/11
- 23.1MCP Client Initialization
- 23.2Server Capability Discovery
- 23.3Dynamic Tool Discovery
- 23.4MCP-to-LLM Tool Mapping
- 23.5Argument Translation
- 23.6Result Translation
- 23.7Multiple MCP Servers
- 23.8Tool Namespacing
- 23.9Transport Errors
- 23.10Connection Lifecycle
- 23.11Agent and MCP Integration Testing
0/13
- 24.1Authentication and Authorization
- 24.2OAuth Fundamentals
- 24.3Token Validation
- 24.4Scope Enforcement
- 24.5User Identity Propagation
- 24.6Local Process Trust Boundaries
- 24.7TLS and Transport Security
- 24.8SSRF Protection
- 24.9Prompt Injection Through MCP
- 24.10Tool Allowlists
- 24.11Rate Limiting
- 24.12Sensitive Data Filtering
- 24.13Audit Logging
Machine Learning Engineering
0 / 2220/13
- 25.1Python Syntax and Data Types
- 25.2Functions and Classes
- 25.3Collections
- 25.4Iterators and Generators
- 25.5Comprehensions
- 25.6Exceptions and File Handling
- 25.7Modules and Packages
- 25.8Type Hints and Data Classes
- 25.9Virtual Environments
- 25.10Jupyter Notebooks
- 25.11Dependency Management
- 25.12Testing with pytest
- 25.13Project Structure
0/17
- 26.1NumPy Arrays
- 26.2Shapes and Dimensions
- 26.3Data Types
- 26.4Vectorization
- 26.5Broadcasting
- 26.6Indexing and Slicing
- 26.7Matrix Operations
- 26.8Pandas Series and DataFrames
- 26.9File Ingestion
- 26.10Filtering and Sorting
- 26.11Grouping and Aggregation
- 26.12Joins and Merges
- 26.13Missing Values
- 26.14Duplicate Records
- 26.15Date and Time Processing
- 26.16Exploratory Data Analysis
- 26.17Memory-Efficient Processing
0/14
- 27.1Population and Sample
- 27.2Mean, Median and Mode
- 27.3Variance and Standard Deviation
- 27.4Percentiles and Quantiles
- 27.5Distribution Shape
- 27.6Outliers
- 27.7Sampling and Sampling Bias
- 27.8Correlation and Causation
- 27.9Covariance
- 27.10Confidence Intervals
- 27.11Hypothesis Testing
- 27.12p-Values
- 27.13Statistical Significance
- 27.14Type I and Type II Errors
0/16
- 28.1Sample Spaces and Events
- 28.2Conditional Probability
- 28.3Independent Events
- 28.4Bayes' Theorem
- 28.5Random Variables
- 28.6Discrete Distributions
- 28.7Continuous Distributions
- 28.8Bernoulli Distribution
- 28.9Binomial Distribution
- 28.10Normal Distribution
- 28.11Expected Value
- 28.12Variance
- 28.13Likelihood
- 28.14Maximum Likelihood Estimation
- 28.15Law of Large Numbers
- 28.16Central Limit Theorem
0/13
- 29.1Scalars, Vectors and Matrices
- 29.2Vector Operations
- 29.3Dot Products
- 29.4Matrix Multiplication
- 29.5Vector Norms and Distances
- 29.6Linear Independence
- 29.7Basis and Rank
- 29.8Matrix Inverse
- 29.9Eigenvalues and Eigenvectors
- 29.10Orthogonality
- 29.11Projections
- 29.12Singular Value Decomposition
- 29.13Dimensionality Reduction
0/15
- 30.1Features and Target Variables
- 30.2Data Types and Validation
- 30.3Missing-Value Imputation
- 30.4Duplicate and Inconsistent Data
- 30.5Outlier Detection
- 30.6Categorical Encoding
- 30.7One-Hot Encoding
- 30.8Ordinal Encoding
- 30.9Standardization
- 30.10Normalization
- 30.11Class Imbalance
- 30.12Train/Validation/Test Splits
- 30.13Data Leakage
- 30.14Time-Based Splitting
- 30.15Reproducible Preprocessing
0/13
- 31.1Regression Fundamentals
- 31.2Simple Linear Regression
- 31.3Multiple Linear Regression
- 31.4Model Assumptions
- 31.5Loss Functions
- 31.6Mean Squared Error
- 31.7Mean Absolute Error
- 31.8Gradient Descent
- 31.9Learning Rate and Convergence
- 31.10Ridge Regression
- 31.11Lasso Regression
- 31.12Underfitting and Overfitting
- 31.13Residual Analysis
0/15
- 32.1Binary Classification
- 32.2Multiclass Classification
- 32.3Logistic Regression
- 32.4Sigmoid Function
- 32.5Softmax Function
- 32.6Decision Boundaries
- 32.7Confusion Matrix
- 32.8Precision and Recall
- 32.9F1-Score
- 32.10ROC-AUC
- 32.11PR-AUC
- 32.12Threshold Selection
- 32.13Probability Calibration
- 32.14Class Weights
- 32.15Cost-Sensitive Classification
0/13
- 33.1Decision Tree Fundamentals
- 33.2Tree Structure
- 33.3Gini Impurity
- 33.4Entropy
- 33.5Information Gain
- 33.6Splitting Criteria
- 33.7Stopping Conditions
- 33.8Tree Overfitting
- 33.9Bagging
- 33.10Random Feature Selection
- 33.11Random Forest Training
- 33.12Feature Importance
- 33.13Bias-Variance Trade-off
0/13
- 34.1Boosting vs. Bagging
- 34.2Sequential Error Correction
- 34.3Gradient Boosting Fundamentals
- 34.4Loss Functions and Residuals
- 34.5Learning Rate
- 34.6Number of Estimators
- 34.7Tree Depth and Regularization
- 34.8XGBoost
- 34.9LightGBM
- 34.10CatBoost
- 34.11Categorical Feature Handling
- 34.12Early Stopping
- 34.13Model Benchmarking
0/10
- 35.1Supervised vs. Unsupervised Learning
- 35.2K-Means
- 35.3Centroids and Distance Metrics
- 35.4Selecting the Number of Clusters
- 35.5Feature Scaling
- 35.6Hierarchical Clustering
- 35.7DBSCAN
- 35.8Silhouette Score
- 35.9Cluster Interpretation
- 35.10Dimensionality Reduction for Visualization
0/13
- 36.1Domain-Driven Features
- 36.2Numerical Transformations
- 36.3Categorical Features
- 36.4Interaction Features
- 36.5Date and Time Features
- 36.6Aggregation Features
- 36.7Window Features
- 36.8Missingness Indicators
- 36.9Feature Selection
- 36.10Multicollinearity
- 36.11Leakage-Resistant Features
- 36.12Training-Serving Consistency
- 36.13Feature Stores
0/13
- 37.1Baseline Models
- 37.2Metric Selection
- 37.3Training and Validation
- 37.4Regression Metrics
- 37.5Classification Metrics
- 37.6Threshold Evaluation
- 37.7Calibration Evaluation
- 37.8Error Analysis
- 37.9Subgroup Performance
- 37.10Generalization
- 37.11Robustness
- 37.12Business Impact
- 37.13Reproducible Evaluation
0/8
- 38.1K-Fold Cross-Validation
- 38.2Stratified K-Fold
- 38.3Leave-One-Out
- 38.4Group-Based Splitting
- 38.5Time-Series Cross-Validation
- 38.6Leakage Across Folds
- 38.7Cross-Validation Limitations
- 38.8Evaluation Variance
0/10
- 39.1Parameters vs. Hyperparameters
- 39.2Manual Search
- 39.3Grid Search
- 39.4Random Search
- 39.5Bayesian Optimization
- 39.6Search Spaces
- 39.7Validation-Based Selection
- 39.8Early Stopping
- 39.9Experiment Tracking
- 39.10Reproducibility and Compute Budgets
0/12
- 40.1Training vs. Inference
- 40.2Scikit-Learn Transformers
- 40.3Estimators
- 40.4Pipeline API
- 40.5ColumnTransformer
- 40.6Preprocessing and Model Integration
- 40.7Pipeline Serialization
- 40.8Training Data Versioning
- 40.9Experiment Tracking
- 40.10Model Artifact Versioning
- 40.11Automated Training Tests
- 40.12Training-Serving Skew Prevention
0/14
- 41.1Batch Inference
- 41.2Online Inference
- 41.3REST Inference APIs
- 41.4FastAPI Fundamentals
- 41.5Model Loading and Serialization
- 41.6Input Validation
- 41.7Response Schemas
- 41.8Model Versioning
- 41.9Cold Starts
- 41.10CPU vs. GPU Inference
- 41.11Dockerized Model Services
- 41.12Spring Boot–Python Integration
- 41.13Latency and Throughput
- 41.14Prediction Monitoring
Production AI/ML Engineering
0 / 1240/12
- 42.1AI Threat Modeling
- 42.2Prompt Injection
- 42.3Sensitive Data Exposure
- 42.4Retrieval Access Control
- 42.5Tenant Isolation
- 42.6Tool Permission Boundaries
- 42.7Insecure Model Outputs
- 42.8Dependency Security
- 42.9Secrets Management
- 42.10Audit Logging
- 42.11Red-Team Testing
- 42.12OWASP Guidance for LLM Applications
0/12
- 43.1Input Validation
- 43.2Output Validation
- 43.3Structured Output Enforcement
- 43.4Content and Policy Checks
- 43.5Refusal Handling
- 43.6Grounded-Answer Requirements
- 43.7Citation Verification
- 43.8PII Detection and Redaction
- 43.9Tool Allowlists
- 43.10Human Approval Gates
- 43.11Ambiguous Request Handling
- 43.12Guardrail Evaluation
0/12
- 44.1Versioned Golden Datasets
- 44.2Automated Offline Evaluation
- 44.3CI Regression Tests
- 44.4Retrieval Evaluation
- 44.5Answer Quality Evaluation
- 44.6Tool-Calling Evaluation
- 44.7LLM-as-Judge Calibration
- 44.8Human Review
- 44.9Adversarial Test Suites
- 44.10Model and Prompt Comparison
- 44.11Release Quality Gates
- 44.12Production Feedback and Trace Replay
0/13
- 45.1Request Volume
- 45.2Error Rates
- 45.3Latency Percentiles
- 45.4Token Usage and Cost
- 45.5Model Availability
- 45.6Prediction Distributions
- 45.7Feature Statistics
- 45.8Retrieval Quality
- 45.9Groundedness Monitoring
- 45.10Tool-Call Failures
- 45.11Dashboards and Alerts
- 45.12Logs, Metrics and Traces
- 45.13Incident Diagnosis
0/12
- 46.1Data Drift
- 46.2Concept Drift
- 46.3Training-Serving Skew
- 46.4Feature Distribution Changes
- 46.5Prediction Distribution Changes
- 46.6Statistical Drift Tests
- 46.7Population Stability Index
- 46.8Ground-Truth Performance Monitoring
- 46.9Delayed Labels
- 46.10Drift Alerting
- 46.11Retraining Triggers
- 46.12Model Rollback
0/14
- 47.1End-to-End Latency Budgets
- 47.2Model Selection
- 47.3Token Optimization
- 47.4Context Reduction
- 47.5Retrieval Optimization
- 47.6Prompt Optimization
- 47.7Parallel Tool Execution
- 47.8Streaming Responses
- 47.9Batch Inference
- 47.10Timeouts and Deadlines
- 47.11Retry Costs
- 47.12Quality-Cost Trade-offs
- 47.13Load Testing
- 47.14Capacity Planning
0/14
- 48.1Application Caching
- 48.2Model Caching
- 48.3Exact-Response Caching
- 48.4Semantic Caching
- 48.5Embedding Caching
- 48.6Retrieval-Result Caching
- 48.7Cache Key Design
- 48.8Tenant Isolation
- 48.9TTL and Invalidation
- 48.10Version-Aware Cache Keys
- 48.11Stale-Data Risks
- 48.12Cache Stampedes
- 48.13Cache Hit Ratio
- 48.14Cache Performance Evaluation
0/16
- 49.1Timeouts and Cancellation
- 49.2Exponential Backoff and Jitter
- 49.3Circuit Breakers
- 49.4Bulkheads
- 49.5Concurrency Limits
- 49.6Rate Limits and Quotas
- 49.7Idempotency
- 49.8Graceful Degradation
- 49.9Fallback Strategies
- 49.10Partial Results
- 49.11Queue-Based Processing
- 49.12Dead-Letter Queues
- 49.13Dependency Failures
- 49.14Health and Readiness Checks
- 49.15Recovery Objectives
- 49.16Fault Injection
0/19
- 50.1Environment Configuration
- 50.2Secrets Management
- 50.3Docker Images
- 50.4Multi-Stage Builds
- 50.5CI/CD Pipelines
- 50.6Unit and Integration Testing
- 50.7Evaluation Tests in CI/CD
- 50.8Database Migrations
- 50.9Staging and Production Environments
- 50.10Health and Readiness Probes
- 50.11Model Artifact Storage
- 50.12Model and Prompt Versioning
- 50.13Canary Deployments
- 50.14Rolling Deployments
- 50.15Rollback Strategies
- 50.16Autoscaling and Resource Limits
- 50.17Production Observability
- 50.18Cost Controls
- 50.19Operational Runbooks