12 skill groups · 7 courses · 702 goals

GenAI Safety & Evaluation Engineering

Design automated LLM evaluation pipelines, red-team GenAI systems, build bias detection and fairness benchmarks, implement guardrails.

Preview 7 goals free Subscribe

cancel anytime · save 20% on 6 months

7 courses, each building on the last. The bar shows each course's share of the curriculum. 7 goals are unlocked for preview.

Python Essentials for Agent Builders9.5%62 goals

Your Dev Environment4 goalsNavigate filesystem with terminal · Manage files from command line · Set up VS Code · Configure terminal in VS Code

Python, Git & Package Management6 goalsInstall and verify Python · Write hello world script · Use Python REPL · Initialize Git repository · Track changes with Git · Install packages with pip

Variables & Basic Types5 goalsCreate and name variables · Work with strings · Work with numbers · Work with booleans · Format with f-strings

Control Flow4 goalsMake decisions with if/elif/else · Iterate with for loops · Repeat with while loops · Control loop execution

Functions5 goalsDefine and call functions · Use parameters · Return values · Document with docstrings · Understand scope

Modules & Imports4 goalsImport standard library · Create custom modules · Understand Python path · Create packages

Lists & Tuples5 goalsCreate and access lists · Modify lists · Slice lists · Use list comprehensions · Work with tuples

Dictionaries & Sets5 goalsCreate and access dicts · Modify dictionaries · Iterate over dicts · Work with nested dicts · Use sets

Classes & Dataclasses5 goalsUnderstand class basics · Create dataclasses · Add methods · Use default values · Basic inheritance

Files, JSON & Error Handling5 goalsRead and write files · Work with JSON · Use pathlib · Handle exceptions · Create custom exceptions

Basic Testing4 goalsUse assert statements · Create test functions · Run pytest · Test classes

Environment Variables & Configuration5 goalsUnderstand environment variables · Use .env files · Load with python-dotenv · Handle missing variables · Organize configuration

Decorators & Context Managers5 goalsUnderstand decorators · Write simple decorators · Use context managers · Write context managers · Combine patterns

LLM Foundations for Agent Builders7.6%50 goals

Generators & Iterators5 goalsUnderstand iteration · Create generators · Use generator expressions · Build data pipelines · Use itertools

Async Programming Basics5 goalsUnderstand async concepts · Write async functions · Run concurrent operations · Use async context managers · Handle async exceptions

Type Hints & Pydantic5 goalsAdd basic type hints · Use typing generics · Create Pydantic models · Validate API data · Configure Pydantic

Data Pipelines & Transformations5 goalsBuild functional pipelines · Work with tabular data · Transform data shapes · Process LLM data formats · Optimize for performance

HTTP Clients & httpx5 goalsMake GET requests · Make POST requests · Use async httpx · Handle errors · Use sessions

Your First LLM Call5 goalsSet up credentials · Install Gemini SDK · Make first API call · Parse response · Handle API errors

Sampling Parameters & Output Control5 goalsUnderstand temperature · Use top-p sampling · Implement determinism · Control output length · Use structured output

Multi-Provider & Prompt Engineering5 goalsBuild provider abstraction · Structure conversations · Use few-shot prompting · Implement chain-of-thought · Build prompt templates

Cost Awareness & Token Economics5 goalsUnderstand pricing models · Calculate request costs · Compare provider costs · Identify cost drivers · Basic cost optimization

Retry Patterns with Tenacity5 goalsUnderstand retry need · Use tenacity basics · Implement exponential backoff · Handle specific exceptions · Combine with async

Kubernetes Essentials for GenAI Engineers7.3%48 goals

Containerizing LLM Applications6 goalsWrite a Python app that calls the Gemini API and returns structured responses · Write a Dockerfile and build a container image for the LLM app · Run the containerized LLM app with environment-based configuration · Use Docker Compose to run the LLM app with supporting services · Tag images with semantic versions and push to a container registry · Debug containers with exec, logs, and inspect

Your Kubernetes Cluster & First LLM Pod6 goalsUnderstand K8s architecture and connect to your vCluster · Deploy the LLM app as your first Kubernetes pod · Organize workloads with namespaces · Use labels and selectors to organize and query resources · Understand pod lifecycle and restart policies · Master kubectl debugging: exec, logs, describe, port-forward

Services & the LLM Chat Backend6 goalsCreate a ClusterIP service to expose the LLM chat API internally · Deploy a multi-tier LLM chat application · Compare service types: ClusterIP, NodePort, LoadBalancer · Master DNS-based service discovery in Kubernetes · Understand endpoints and traffic routing · Debug service connectivity problems

Deployments, Scaling & Rolling Updates6 goalsCreate a Deployment for the LLM chat API · Scale LLM app replicas to handle concurrent requests · Perform a rolling update with zero downtime · Roll back a broken deployment · Compare deployment strategies: RollingUpdate vs Recreate · Manage deployment lifecycle with kubectl rollout

Packaging with Helm & Kustomize6 goalsCreate a Helm chart for the LLM chat application · Parameterize the chart with values.yaml for each environment · Manage Helm release lifecycle: install, upgrade, rollback · Use Kustomize bases and overlays for the LLM app · Use Kustomize patches and generators · Compare Helm vs Kustomize for different deployment scenarios

Networking, Ingress & TLS6 goalsExpose the LLM chat API via an Ingress resource · Add TLS to the Ingress for HTTPS access · Isolate services with NetworkPolicies · Configure Ingress annotations for production traffic · Understand K8s networking: pod IPs, CNI, and service routing · Debug networking and connectivity issues

Health Probes, Autoscaling & Self-Healing6 goalsAdd liveness and readiness probes to the LLM chat API · Configure startup probes for containers with slow initialization · Scale the chat API automatically with HPA based on CPU · Create PodDisruptionBudgets for safe maintenance · Implement health check patterns for LLM-dependent services · Combine autoscaling, probes, and PDBs for a resilient LLM service

RBAC, Security & K8s Troubleshooting6 goalsCreate RBAC roles for the LLM chat application · Enforce Pod Security Standards · Apply SecurityContext for defense in depth · Debug CrashLoopBackOff and OOMKilled failures · Use kubectl debug and ephemeral containers for live debugging · Troubleshoot LLM-specific issues: timeouts, proxy errors, stale connections

Web APIs & Services for GenAI Engineers9.1%60 goals

FastAPI Fundamentals6 goalsCreate a FastAPI application with path operations · Define Pydantic request and response models · Implement dependency injection for shared resources · Build CRUD endpoints with proper HTTP semantics · Configure OpenAPI documentation with examples · Handle errors with custom exception handlers

Async Python for APIs6 goalsConvert sync endpoints to async with proper await patterns · Implement background tasks for non-blocking operations · Execute concurrent API calls with asyncio.gather · Manage application lifecycle with lifespan handlers · Build async generators for streaming responses · Control concurrency with semaphores and throttling

Database Integration6 goalsConfigure SQLAlchemy async engine with connection pooling · Define ORM models with relationships and constraints · Create and manage database migrations with Alembic · Implement repository pattern for data access · Build transactional endpoints with session lifecycle · Implement filtering, sorting, and full-text search

Authentication & Authorization6 goalsImplement user registration with password hashing · Build OAuth2 password flow with JWT tokens · Implement API key authentication for services · Enforce role-based access control with permissions · Build token refresh and revocation · Compose multiple auth strategies into dependencies

Real-time Streaming6 goalsBuild SSE endpoint for streaming LLM responses · Implement WebSocket endpoint with connection lifecycle · Build WebSocket connection manager for broadcasting · Handle backpressure and slow clients · Implement heartbeat and automatic reconnection · Build real-time notification system with Redis pub/sub

Resilience Patterns6 goalsImplement rate limiting with Redis sliding window · Build circuit breaker for LLM provider calls · Configure retry logic with tenacity · Isolate critical paths with bulkhead semaphores · Build fallback responses for degraded mode · Combine resilience patterns into middleware stack

API Gateway & Routing6 goalsBuild reverse proxy with path-based routing · Implement load balancing across backend instances · Transform requests and responses through the gateway · Aggregate responses from multiple backends · Implement service discovery with health checking · Build gateway authentication and request enrichment

Testing & Documentation6 goalsWrite async endpoint tests with httpx.AsyncClient · Build database fixtures with transaction rollback · Mock external services for deterministic tests · Implement contract tests for API consumers · Measure test coverage and set quality gates · Generate rich OpenAPI documentation with examples

API Versioning & Evolution6 goalsImplement URL-based API versioning with routers · Build header-based version negotiation · Manage deprecation with Sunset and Warning headers · Build request and response adapters for version translation · Detect breaking changes automatically · Generate API changelogs from schema diffs

Deployment & Observability6 goalsBuild production Docker images with multi-stage builds · Deploy to Kubernetes with health check probes · Instrument endpoints with Prometheus metrics · Implement distributed tracing with OpenTelemetry · Build structured logging with correlation IDs · Create Grafana dashboards for API monitoring

GenAI Agent Engineering35.4%232 goals

The LLM Client7 goalsOpenAI client setup · Anthropic client setup · Google Gemini client setup · Build a unified LLM client interface · Error handling and provider fallback · Async LLM client patterns · Practical use cases — security, parameters, observability

Token Economics7 goalsUnderstand tokenization · Count tokens across providers · Cost forecasting and budgeting · Track LLM API usage in production · Implement budget controls · Optimize tokens · Advanced context engineering

Prompt Caching4 goalsImplement Anthropic cache_control · Leverage OpenAI automatic caching · Design cache-friendly prompt architectures · Build cache monitoring systems

The Function Caller7 goalsOpenAI function schemas · Anthropic function schemas · Gemini function schemas · Handle tool call responses · Execute tools safely with Pydantic validation · Handle parallel tool calls · Framework integration with LangGraph

The Tool Definer7 goalsWrite clear tool descriptions for LLMs · Define parameter schemas · Use Pydantic for tool schemas · Implement tool decorators · Handle complex parameter types · Validate tool inputs at runtime · Framework tool patterns — LangGraph, CrewAI, OpenAI, Gemini, Anthropic

The Raw Agent Loop7 goalsThe core agent while-loop · Manage context as a mutable list · Handle stop sequences · Track iteration limits · Tool execution in the loop · Build a conversation state tracker · Build with LangGraph StateGraph

The Prompt Engineer (Dynamic)6 goalsMaster Jinja2 templating for prompts · Implement dynamic few-shot example selection · Enforce Chain-of-Thought reasoning · Structure system prompts with a builder pattern · Inject dynamic context into prompts safely · Build prompt versioning and A/B testing

The ReAct Pattern (Manual)6 goalsBuild the Thought-Action generator · Tool execution and observation injection · Complete ReAct agent implementation · Advanced ReAct patterns — validation, retry, confidence · Optimize ReAct performance · Common ReAct pitfalls and solutions

The Planner Pattern7 goalsPlan generation · Step execution · Dynamic replanning · Hierarchical planning · Plan optimization · Monitoring and observability · Practical considerations — strategy selection

The Pydantic Tool7 goalsPydantic fundamentals for tool definitions · Generate JSON Schema from Pydantic models · Input validation with custom validators · Build a Pydantic tool library · Advanced Pydantic patterns · Integrate Pydantic tools with agent frameworks · Common pitfalls and solutions

The Safe Executor (Sandboxing)5 goalsUnderstand code execution risks · Static code analysis · Sandboxed execution · Apply resource limits · Build a complete safe executor

The Web Navigator5 goalsWeb navigation fundamentals · Web navigation tools — locating elements and forms · Browser automation with Playwright · Session management · Complete web navigator system

The MCP Protocol (Basics)4 goalsJSON-RPC 2.0 message format and handler · Transport mechanisms — stdio and HTTP/SSE · Protocol lifecycle — initialization, runtime, shutdown · Capability negotiation

The MCP Server6 goalsCreate an MCP server with lifecycle management · Define MCP tools · Implement MCP resources · Create prompt templates · Error handling in MCP servers · Composable MCP server architecture

The MCP Client6 goalsMCP client architecture and stdio transport · Discover available tools and translate schemas · Proxy tool invocation · Fetch and use MCP resources · Manage MCP server lifecycle · Build multi-server MCP clients

The Tool Router5 goalsTool routing architecture and implementation · Namespace-based routing · Capability-based routing · Fallback chains · Routing performance optimization

Short-Term Memory8 goalsSliding window memory · Token-aware memory management · Message summarization strategies · Memory persistence layers · Memory retrieval optimization · Integrate memory with agents · Memory performance considerations · Non-functional requirements (privacy + safety)

Long-Term Memory (RAG)6 goalsDocument chunking strategies · Embedding pipelines · Vector database integration · Hybrid search implementation · Retrieval optimization · RAG response generation

Agentic RAG Patterns5 goalsSelf-reflective RAG · Multi-hop retrieval · Query routing · Adaptive retrieval · Retrieval feedback loops

Semantic Memory6 goalsKnowledge extraction pipelines · Entity and relationship extraction · Knowledge graph construction · Memory consolidation · Integrate semantic memory with agents · Build semantic memory with LangGraph

Context Optimizer6 goalsContext economics · Dynamic context prioritization · Context compression techniques · Prompt optimization · Context utilization metrics · Complete context optimizer

The State Graph5 goalsStateGraph fundamentals — config and lifecycle · Design state schemas with TypedDict · Add nodes to StateGraph · State initialization patterns · Tracing, debugging, validation

The Conditional Edge5 goalsUnderstand conditional edges · Design routing functions · Fan-out and fan-in patterns · Handle unknown routes and errors · Multi-stage routing

The Checkpointer (Time Travel)4 goalsResumable workflows · Inspect, replay, and time-travel · Retention, large state, and performance · Thread management — IDs and namespaces

Human-in-the-Loop6 goalsLangGraph interrupt patterns · Approval workflow patterns · Interactive agent conversations · Feedback integration · State management for HITL · Practical use cases — escalation and analytics

The Streaming Agent6 goalsStreaming modes in LangGraph · Token streaming from LLMs · Custom events with `astream_events` · Build streaming APIs · Error handling in streams · Backpressure and flow control

The Subgraph (Composition)7 goalsSubgraph fundamentals — compile + test in isolation · State schema mapping · Subgraph checkpointers + namespace isolation · Compose subgraphs into a parent · Catch subgraph exceptions and recover · Define subgraph interfaces and build a registry · Build a multi-agent orchestrator

The Supervisor Pattern7 goalsDesign supervisor architectures · Worker agent specialization · Build the complete supervisor graph · Manage inter-agent communication · Handle failures and edge cases · Implement task aggregation · Build the supervisor pattern with CrewAI

The Hierarchical Pattern4 goalsDesign hierarchical agent architectures · Implement team-lead agents · Build cross-team coordination · Build the complete hierarchical graph

The Reflector Pattern (Critique)6 goalsDesign reflection architectures · Implement critic agents · Build the evaluation and convergence system · Build the complete reflection graph · Handle reflection edge cases · Practical use cases for reflection

Input Guardrails6 goalsDesign layered guardrail architectures · Format and schema validation · Build content filtering systems · Create injection / jailbreak detection · Implement policy-based guardrails · Assemble the complete guardrail system

Output Guardrails6 goalsDesign output validation architectures · Implement factual validation (hallucination detection) · Build content safety filters · Create PII redaction · Implement policy compliance · Assemble the complete output guardrail system

Prompt Injection Defense7 goalsIdentify injection vulnerabilities · Detect direct injections · Detect indirect injections · Implement defense layers · Build red-team suites · Implement canary tokens · LangGraph injection defense pipeline

Evaluations (Evals)6 goalsDesign evaluation frameworks · Implement automated evaluation pipelines · Create task-specific metrics · Human evaluation protocols · Regression testing · Set baselines and track progress

Agent Benchmarking6 goalsUnderstand the GAIA benchmark · Implement ToolBench evaluation · Use AgentBench · Design domain-specific benchmarks · Cross-model performance comparison · Build benchmark dashboards

Tracing & Observability6 goalsUnderstand distributed tracing · Add tags and metadata · Context propagation · Build feedback collection · Integrate with Langfuse · Trace visualization

Tool Use Debugging6 goalsTool selection failures and solutions · Argument validation systems · Build tool use dashboards and visualization · Schema mismatch detection · Tool call replay · Interactive tool debugger

Agent Safety Boundaries6 goalsTool permission systems · Resource budget limiters · Kill switch mechanisms · Sandbox isolation · Safety monitoring and escalation · Safety boundaries integration — capstone

Autonomous Agent Governance6 goalsImmutable audit trails · Decision logging framework · Human escalation engine · Compliance report generator · Governance middleware · Governance capstone

GenAI Evaluation, Safety & Governance22.9%150 goals

Evaluation Dataset Curation6 goalsBuild a stratified evaluation dataset · Implement dataset versioning and snapshots · Detect dataset contamination and leakage · Build automated dataset refresh pipeline · Create dataset cards for documentation · Build a dataset annotation pipeline

LLM-as-Judge Evaluation6 goalsBuild a rubric-based LLM judge · Implement pairwise comparison evaluation · Calibrate judge against human scores · Detect and mitigate judge biases · Build multi-judge consensus with Vertex AI Evaluation Service and MLflow MemAlign · Build a judge evaluation API service

RAG Evaluation with RAGAS & DeepEval6 goalsImplement RAGAS metrics for RAG evaluation · Build DeepEval test suites for RAG · Evaluate retrieval quality independently · Build end-to-end RAG evaluation pipeline · Compare RAG configurations with statistical testing · Integrate RAG evaluation into CI/CD

Evaluation Observability with Langfuse v3 & OpenTelemetry6 goalsDeploy Langfuse v3 on GKE with OpenTelemetry-native instrumentation · Build evaluation dashboards in Langfuse · Track costs and token usage across providers · Manage prompt versions with Langfuse · Compare observability platforms: Langfuse vs Arize Phoenix vs Braintrust · Build automated evaluation alerting

Agent Trajectory Evaluation6 goalsScore agent tool selection with DeepEval 3.0 and Vertex AI Agent Evaluation · Evaluate agent quality with Patronus AI Percival agent-as-a-judge · Implement process-oriented vs outcome evaluation · Build agent benchmarks with task suites · Evaluate error recovery and self-correction · Track agent evaluation in production with Langfuse

Human-in-the-Loop Evaluation6 goalsDeploy Argilla and create annotation projects · Build preference data collection for RLHF/DPO · Implement annotation quality control · Build automated pre-annotation with LLM suggestions · Export annotations for model feedback loops · Run an end-to-end annotation campaign

A/B Testing for LLM Systems6 goalsDesign A/B experiments for prompt variants · Build traffic splitting with consistent assignment · Implement statistical significance testing · Monitor experiments with guardrail metrics · Run a model swap experiment (OpenAI vs Gemini) · Build experiment results dashboard

Evaluation-Driven CI/CD & Continuous Production Monitoring6 goalsBuild evaluation gates with Promptfoo and DeepEval in CI · Implement progressive evaluation tiers · Implement cost-aware evaluation budgets · Build release validation pipeline · Track evaluation trends across releases · Build continuous production monitoring with async scoring

Cross-Model Evaluation6 goalsBuild a standardized multi-provider eval harness · Compare structured output compliance across providers · Build cost-performance analysis across providers · Test provider reliability and error handling · Build a model selection decision matrix · Implement model migration playbook

Cost Governance & Token Budgets6 goalsBuild per-user token usage tracking · Implement budget enforcement and rate limiting · Detect cost anomalies and spending spikes · Build intelligent model routing with LiteLLM gateway and RouteLLM semantic routing · Implement prompt compression and caching · Build cost governance dashboard and chargeback

Prompt Injection Defense6 goalsDetect prompt injection with PromptGuard 2 and custom classifiers · Defend against indirect prompt injection · Prevent system prompt leakage (OWASP LLM07) · Build canary token detection · Deploy LlamaFirewall and Google Model Armor as unified defense orchestrators · Implement output filtering and response safety

Content Safety Filters6 goalsCompare guardrail frameworks: Guardrails AI vs NeMo Guardrails 0.20 vs NemoGuard NIMs vs Google Model Armor · Build custom domain-specific safety validators · Integrate LlamaGuard 4 and hosted LLM safety APIs · Build multi-layer content safety pipeline · Tune filter sensitivity and manage false positives · Monitor content safety metrics in production

PII Detection & Redaction6 goalsDetect PII with Presidio and Google Sensitive Data Protection · Implement reversible PII redaction · Build custom PII recognizers for domain data · Implement PII audit logging for compliance · Scan LLM outputs for PII leakage · Build end-to-end PII protection pipeline

Hallucination Detection6 goalsDetect hallucinations with source grounding and Patronus AI Lynx 2.0 · Implement citation verification · Build NLI-based faithfulness scoring · Classify hallucination types · Build production hallucination monitoring · Reduce hallucinations with prompt engineering

Adversarial Robustness Testing6 goalsExecute manual adversarial attack categories · Automated red teaming with PyRIT, Promptfoo Hydra, and Meta GOAT · Vulnerability scanning with NeMo Auditor and Garak 0.14 · Build adversarial CI/CD test suite · Measure defense effectiveness against attacks · Build adversarial robustness dashboard

Agent Safety, MCP Security & Sandboxing6 goalsValidate agent tool calls against permission policies · Secure MCP servers and implement agent gateway patterns · Build human approval for high-risk operations · Detect privilege escalation in agent behavior · Build agent audit trail with GCP SCC Agent Engine Threat Detection · Build agent safety evaluation framework

Multi-Modal Safety6 goalsDetect adversarial image inputs · Build image content safety filters with LlamaGuard 4 · Defend against cross-modal prompt injection · Implement safe multi-modal processing pipeline · Test vision model hallucination in multi-modal context · Monitor multi-modal safety in production

Vector & Embedding Security6 goalsDetect RAG data poisoning attacks · Implement document-level access control for RAG · Verify embedding integrity · Build adversarial embedding defense · Detect data exfiltration via RAG · Build vector store security dashboard

OWASP LLM Top 10 2025 & MITRE ATLAS6 goalsMap OWASP LLM Top 10 2025 to implemented defenses · Implement defenses for Unbounded Consumption (LLM10) · Build MITRE ATLAS threat models · Build OWASP compliance testing with Promptfoo presets and Checks by Google · Implement supply chain security (LLM03) · Create OWASP + ATLAS compliance dashboard

EU AI Act Compliance6 goalsClassify AI systems under EU AI Act risk categories · Implement the Feb 2025 AI literacy requirements · Build technical documentation for GPAI compliance · Implement risk management system · Build human oversight mechanisms · Track EU AI Act enforcement timeline compliance

Compliance Frameworks6 goalsImplement NIST AI RMF Govern and Map functions · Implement NIST AI RMF Measure and Manage functions · Build ISO 42001 AI management system documentation · Create unified governance dashboard with Credo AI Agent Registry · Implement comprehensive audit trail · Build compliance automation and alerting

Red Teaming Methodology6 goalsPlan and scope a red team exercise · Execute manual red team techniques · Combine manual, Meta GOAT automated, and Inspect AI red teaming · Build red team findings reports · Track remediation and verify fixes · Build AI safety scorecard and establish red team cadence

Bias, Fairness & Continuous Monitoring6 goalsDetect bias in hosted LLM outputs · Implement fairness metrics for LLM applications · Build continuous safety monitoring for production · Detect safety drift over time · Build safety incident response workflow · Generate weekly and monthly safety reports

End-to-End Eval, Safety & Governance Pipeline6 goalsBuild evaluation benchmark gate · Build safety testing gate · Build red-team gate with automated adversarial testing · Build compliance evidence gate · Orchestrate the full pipeline with Argo Workflows · Build pipeline dashboard and deploy to production

Enterprise Safety Operations Capstone6 goalsRespond to simulated adversarial attack on production system · Manage safety through a model provider update · Adapt governance controls for regulatory changes · Detect and remediate safety drift in production · Run a safety war room exercise · Build safety operations dashboard and operational runbook

GenAI Operations8.2%54 goals

Eval Gate Pipeline6 goalsBuild Promptfoo Eval Suites for Pre-Promotion Quality Verification · Implement Eval Gates in Argo Workflows That Block Promotion on Failure · Create Golden Test Sets for Regression Detection · Track Eval Pass Rates and Gate Effectiveness Metrics · Implement performance optimization for automated eval gates · Build operational documentation for automated eval gates

GenAI Alert System6 goalsConfigure Alertmanager with GenAI-Specific Routing Rules and Severity Classification · Deploy Grafana OnCall for On-Call Schedules, Escalation Policies, and Incident Lifecycle · Implement Alert Deduplication and Grouping for Noisy GenAI Metrics · Build Alert Effectiveness Tracking to Reduce Alert Fatigue · Implement performance optimization for alerting strategy · Build operational documentation for alerting strategy

Injection Monitoring System6 goalsImplement multi-layer prompt injection detection with pattern and embedding-based methods · Build real-time injection alerting with severity classification · Create injection attack analysis dashboards for security monitoring · Implement adaptive detection that learns from new attack patterns · Implement performance optimization for prompt injection monitoring · Build operational documentation for prompt injection monitoring

Guardrail Operations Platform6 goalsDeploy Guardrails AI and LlamaFirewall on K8s for runtime content validation · Implement hot-reload guardrail configuration without service restarts · Build A/B testing framework for guardrail thresholds to optimize block rates · Build testing and validation for guardrail operations · Implement performance optimization for guardrail operations · Build operational documentation for guardrail operations

Compliance Audit Engine6 goalsImplement automated compliance scans for GenAI-specific requirements · Build evidence collection pipelines that gather audit artifacts · Schedule recurring compliance checks with drift detection · Build testing and validation for compliance audit automation · Implement performance optimization for compliance audit automation · Build operational documentation for compliance audit automation

Red Team Automation Platform6 goalsBuild automated red team attack suites using Promptfoo for systematic security testing · Implement scheduled security testing with regression detection across model changes · Build security posture scoring with trend monitoring and improvement tracking · Build testing and validation for red team operations · Implement performance optimization for red team operations · Build operational documentation for red team operations

Eval Gates in CI/CD6 goalsEval Suite Design · CI Pipeline Integration · Eval Result Storage · Change-Type-Specific Gates · Eval Failure Workflow · Eval Gates Capstone

AI Incident Response6 goalsAI Incident Taxonomy · Automated Detection · AI Incident Runbooks · Post-Incident Review · Escalation Path Design · Incident Response Capstone

AI Governance Compliance Ops6 goalsEU AI Act Controls · NIST AI RMF Implementation · SOC2 AI Controls · Compliance Dashboard · Audit Preparation Workflow · Compliance Ops Capstone

Core responsibilities this discipline prepares you for. Open one to see how you build it.

1Build automated evaluation pipelines to continuously measure LLM output quality

Design evaluation harnesses with RAGAS, DeepEval, and NeMo Evaluator SDK for multi-metric scoring
Create evaluation datasets with ground-truth annotations and run cross-provider comparisons
Wire CI gates that automatically block deployments when faithfulness or relevance scores degrade

2Conduct red-team exercises — probe LLMs for vulnerabilities

Automate adversarial testing with Garak for prompt injection, jailbreak, and data extraction probes
Run multi-turn adversarial campaigns with Meta GOAT and DeepTeam for agent vulnerability testing
Execute red-team campaigns against realistic systems, discover vulnerabilities, and write actionable findings

3Implement production guardrails — content filters, PII detection, jailbreak prevention

Configure NeMo Guardrails with Colang policy language, Llama Guard 4, and Prompt Guard 2
Add Presidio for PII detection/redaction and Model Armor for Google-native content safety
Layer multiple defenses, test against comprehensive attack suites, and quantify safety-vs-helpfulness tradeoffs

4Design GenAI governance frameworks aligned with regulations

Map EU AI Act risk classification and implement NIST AI RMF control frameworks
Build OWASP LLM Top 10 mitigation strategies mapped to technical controls
Create governance artifacts, conduct risk assessments, and build automated audit trail pipelines

5Evaluate GenAI agent behavior — trajectory quality, tool selection accuracy

Build trajectory scoring systems measuring tool selection accuracy and task completion quality
Design human preference alignment tests and regression test suites for agent workflows
Evaluate multi-step agent executions to identify failure modes and build targeted regression tests

6Monitor bias, fairness, and hallucination rates in production

Detect bias across protected attributes using statistical fairness metrics and disparity analysis
Measure hallucination rates through ground-truth comparison and citation verification
Implement continuous bias scanning, hallucination detection, and alerting for metric drift

7Build safety incident response processes for deployed GenAI systems

Design safety monitoring dashboards with severity-based alert routing and escalation paths
Build incident triage workflows with containment procedures and post-incident reporting templates
Simulate safety incidents end-to-end and practice the full detection-to-resolution workflow

8Design LlamaFirewall policies for agent safety

Configure LlamaFirewall middleware for controlling agent tool access and output filtering rules
Set up multi-agent safety boundaries with policy-based execution constraints
Validate firewall policies against adversarial scenarios where agents attempt to bypass controls

Each group becomes a signed node on your public skill graph — a recruiter-auditable record of the labs you actually passed. Share the URL; they see the discipline, the labs, and the signature.

LLM-as-Judge & Scoring

LLM-as-judge rubric design, position-bias correction, calibration against human raters, scoring functions, judge reproducibility, multi-judge ensembling. The core automated-evaluation technique.

Continuous Eval & CI/CD Gates

Automated eval pipelines: eval-driven CI/CD, quality gates that block deploys, regression suites, champion/challenger A/B, prompt-variant testing — plus eval-pipeline observability, tracing (Langfuse/OpenTelemetry) and cost governance for judge runs.

Eval Dataset & Safety Benchmark Design

Golden-set curation, dataset versioning, prompt-variant generation, edge-case mining, stratified sampling — and creating standardized safety benchmarks: jailbreak/refusal taxonomies, harm-category coverage, threshold-gated safety suites.

Adversarial Robustness Evaluation & Red-Teaming

Adversarial robustness evaluation: systematic red-team test-suite generation, jailbreak/injection probe batteries, automated red teaming, robustness scoring and attack-success-rate metrics, OWASP LLM Top 10 + MITRE ATLAS. The measured-verdict eval slice, not offensive exploitation.

Bias, Fairness & Disparate-Impact Benchmarking

Statistical fairness evaluation: demographic-parity and equalized-odds tests, disparate-impact ratios, subgroup and intersectional slicing, counterfactual-token swaps (BBQ/CrowS-Pairs/WinoBias), and fairness-benchmark construction. Distinct from factual grounding.

Compliance & Governance Frameworks

EU AI Act / NIST AI RMF / SOC2 / HIPAA / GDPR control mapping, governance-artifact generation (model cards, eval policies, audit trails, sign-off gates), AI risk classification and governance workflows. Implemented on top of the eval capabilities.

Hallucination & Grounding Eval

Factual-grounding evaluation: hallucination detectors, faithfulness/groundedness scoring, NLI-based entailment, attribution and citation precision against source context. Distinct from statistical fairness.

Agent Trajectory Evaluation

Step-level agent evaluation: trajectory scoring, tool-call accuracy, plan-quality assessment, human-in-the-loop review gates, golden-trajectory datasets.

Content Safety, Guardrails & PII

Runtime safety enforcement: content moderation, guardrails, PII detection + redaction, toxicity classifiers, output sanitization, regulated-content classification. The inference-time defense layer (vs. building the eval suite in G3).

RAG & Cross-Model Evaluation

RAGAS + DeepEval + TruLens pipelines, retrieval relevance + faithfulness + answer-relevancy metrics, cross-model comparison harnesses and arena/pairwise model ranking.

Hosted LLM API Integration

Provider SDK integration in eval and safety code: judge models, multi-provider scoring, cross-model evaluation harnesses, multi-provider abstraction. Prerequisite plumbing.

Python for Eval Engineering

Production-grade Python for eval tooling: async/await, Pydantic models for eval rubrics, typing, dataclasses, pytest harnesses, parametrized testing. Prerequisite.