KnowledgeOS · AI Engineering Academy
Learn AI engineering from first principles to production.
Structured courses for software engineers who want to build, evaluate, and lead enterprise AI systems with real architectural judgment.
- learning options
- 28
- handbook volumes
- 12
- handbook chapters
- 161
- available study
- 218 hr 50 min
Course Catalog
One core path. 27 focused ways to go deeper.
Begin with the complete curriculum, or choose a focused course for the engineering problem directly in front of you.
3 courses
AI Developer Tools
Three practical developer-tool courses: the AI assistants, the coding agents, and the MCP servers that connect them to your systems. Read three complete preview sections free in each, then unlock the full course with a 12-month Course Pass.
AI Developer Tools · Free preview
ChatGPT and Claude for Developers
Use ChatGPT and Claude for real software work and check every answer against a real codebase: prompts that hold up, debugging, tests that catch bugs, SQL and shell, code review, data and research, projects and skills. Then build with the Claude and OpenAI APIs: structured output, tools, retries, caching, cost and evaluations, ending with an acceptance-tested capstone.
Intermediate · professional developers
30 sections · a check for every exercise + an acceptance-tested capstone
About 315 min reading
- ChatGPT and Claude
- Prompting and Verification
- Claude and OpenAI APIs
- Evaluations and Safety
AI Developer Tools · Free preview
Claude Code and Codex: Agentic Coding in Practice
Direct Claude Code and Codex on a real TypeScript app with 14 backlog items and acceptance tests. Plan multi-file changes, fix bugs test-first, prove UI fixes in a browser and review every diff; then add AGENTS.md, skills, subagents, hooks, MCP servers and plugins, and scale out with worktrees, cloud agents, CI and the SDKs.
Intermediate · professional developers
28 sections · 14 backlog items with acceptance tests + capstone
About 275 min reading
- Claude Code
- Codex
- Agent Workflows
- Tests and Review
AI Developer Tools · Free preview
MCP for Developers: Build, Test and Secure Servers
Build a Model Context Protocol server for a real issue tracker in TypeScript and Python. Design tools a model uses correctly, test them over the protocol in both protocol eras, connect them to Claude Code and Codex, serve them over HTTP with tokens, scopes and OAuth metadata, and ship a capstone that asks the person before it changes anything.
Intermediate · professional developers
19 sections · two tested servers + an acceptance-tested capstone
About 295 min reading
- Model Context Protocol
- TypeScript and Python
- Authorization and Security
- Protocol Testing
12-volume curriculum
Core curriculum
Start here for a sequenced path from first principles to enterprise AI architecture.
13 courses
AI Engineering Specializations
Agents, RAG, model adaptation, multimodal systems, serving, security, evaluation, and production delivery.
Software Engineering Track
AI Coding Agents for Professional Engineers
Take coding-agent work from a bounded repository issue to an evidenced patch, then design the security, evaluation, CI/CD, recovery, and operating controls required at enterprise scale.
Intermediate · software engineering
36 sections · 4 parts + capstone
About 150 min reading
- Coding Agents
- Repository Context
- Verification
- Secure Automation
Backend Engineering Track
Production GenAI with Java and Spring AI
Turn Spring engineering skills into production AI systems with typed model boundaries, authorized RAG, guarded tools, MCP, observability, evaluation, resilience, security, and release evidence.
Intermediate to advanced · Java applications
45 sections · 4 parts + engineering studio
About 185 min reading
- Java
- Spring AI 2.0
- Enterprise RAG
- Production Engineering
Applied AI Systems Track
Multimodal AI Engineering
Engineer production systems across images, documents, audio, and video with evidence provenance, multimodal RAG, fusion, accessibility, evaluation, security, cost control, and incident-ready operations.
Intermediate · media applications
45 sections · 4 parts + engineering studio
About 240 min reading
- Vision
- Audio and Video
- Multimodal RAG
- Production AI
AI Security Track
AI Security and Red Teaming for Production Systems
Threat-model and defend production AI across prompts, retrieval, tools, memory, models, supply chains, detection, red-team evidence, containment, and incident recovery.
Advanced · application security
45 sections · 4 parts + security studio
About 245 min reading
- Threat Modeling
- Prompt Injection
- AI Red Teaming
- Incident Response
Model Engineering Track
Fine-Tuning, Alignment and Model Adaptation
Diagnose when weights should change, then engineer governed data, SFT, LoRA and QLoRA, preference optimization, distillation, evaluation, adapter serving, and incident-ready operations.
Advanced · model adaptation
47 sections · 5 parts + adaptation studio
About 275 min reading
- Fine-Tuning
- LoRA and QLoRA
- Preference Optimization
- Adapter Serving
Architecture Track
GenAI System Design and Architecture Interviews
Turn an AI product idea into a quantified production architecture covering capacity, latency, RAG, model routing, tenancy, reliability, evaluation, security, regional recovery, and design interviews.
Advanced · architecture practice
39 sections · 4 parts + capstone
About 155 min reading
- System Design
- Capacity
- Reliability
- Architecture Interviews
Production Engineering Track
LLMOps: Evaluation, Observability and Production Reliability
Turn probabilistic AI behavior into release evidence with representative datasets, calibrated graders, privacy-aware traces, semantic SLOs, controlled canaries, rollback, and incident operations.
Advanced · evaluation and operations
46 sections · 4 parts + reliability studio
About 240 min reading
- LLMOps
- Evaluation
- Observability
- Production Reliability
AI Infrastructure Track
LLM Inference, GPU Serving and Performance Engineering
Turn model artifacts into dependable services through prefill and decode mechanics, KV cache, continuous batching, quantization, GPU topology, distributed serving, SLOs, capacity, cost, and release safety.
Advanced · inference systems
53 sections · 5 parts + inference studio
About 215 min reading
- LLM Inference
- GPU Serving
- Performance Engineering
- Distributed Systems
AI Data Infrastructure Track
AI Data Platform Engineering for Generative AI
Build the governed evidence plane behind RAG, agents, evaluation, and model adaptation through source authority, CDC, document processing, lineage, access control, indexes, deletion, reliability, and cost.
Advanced · data platforms
49 sections · 5 parts + data platform studio
About 260 min reading
- AI Data Platforms
- Streaming and CDC
- Knowledge Pipelines
- Data Governance
ML Systems Track
Distributed AI Training and ML Systems Engineering
Scale a verified training loop through DDP, FSDP, ZeRO, model parallelism, high-throughput data, atomic checkpoints, GPU scheduling, observability, recovery, and outcome-based economics.
Advanced · distributed training
53 sections · 5 parts + training systems studio
About 290 min reading
- Distributed Training
- GPU Clusters
- Parallelism
- ML Systems
Conversational Systems Track
Real-Time Voice AI and Conversational Agent Engineering
Engineer low-latency voice sessions across WebRTC, SIP, streaming speech, turn-taking, interruption, tools, RAG, handoff, evaluation, consent, safety, scaling, and incident operations.
Advanced · conversational systems
49 sections · 5 parts + voice systems studio
About 290 min reading
- Voice AI
- WebRTC and SIP
- Conversational Agents
- Real-Time Systems
Complete Field Manual
AI Agents, Agentic AI & RAG
A continuous path from agent fundamentals to governed RAG, retrieval evaluation, MCP, multi-agent design, and three executable architecture labs.
Intermediate · agents and retrieval
48 sections · 4 parts + capstone
About 500 min reading
- Agents
- Agentic RAG
- Retrieval
- MCP
Hands-on Companion
Agentic AI: The Practitioner’s Companion
Move from concepts to shipping with context budgeting, tool security, durable execution, framework selection, evaluation gates, and three executable labs.
Intermediate · agent implementation
26 sections · 4 parts + studio
About 130 min reading
- Context Engineering
- Agent Loops
- Tooling
- Security
10 courses
Systems Engineering
A ten-course spine for distributed systems, data, cloud, platforms, observability, and architecture decisions.
Systems Engineering Track
Distributed Systems for Production Engineers
Trace lost replies, duplicate charges, slow requests and stale reads through an order and payment case study. Run independent models of retries, tail latency and replication lag, then defend the guarantees and costs of each repair.
Advanced · production systems
35 sections · 2 parts + systems studio
About 165 min reading
- Distributed Systems
- Idempotency
- Consistency and CAP
- Replication
Systems Engineering Track
Sharding, Time and Coordination
Split data across machines without creating hotspots, measure what growth costs in bytes moved, and coordinate work when no two clocks agree and a paused process still believes it holds the lock.
Advanced · production systems
30 sections · 2 parts + coordination studio
About 140 min reading
- Partitioning
- Consistent Hashing
- Clocks and Ordering
- Distributed Locks
Systems Engineering Track
Consensus, Transactions and Resilience
Trace Paxos and Raft from durable state to quorum decisions, move work between systems that share no transaction, and stop one slow dependency from exhausting every thread.
Advanced · production systems
45 sections · 3 parts + resilience studio
About 185 min reading
- Paxos and Raft
- Outbox and Sagas
- Delivery Semantics
- Resilience Patterns
Systems Engineering Track
Databases at Scale
Price a key choice in bytes written, reproduce the isolation anomalies your default level still permits, read the plan that misled the optimiser, and sequence a schema change that stays reversible until the last step.
Advanced · production systems
40 sections · 4 parts + storage studio
About 175 min reading
- Storage Engines
- MVCC and Isolation
- Query Planning
- Caching and Migration
Systems Engineering Track
Kafka and Event-Driven Systems
Measure the exact scope of an ordering guarantee, find the consumer count past which more machines do nothing, place your side effect relative to what exactly-once actually covers, and evolve a schema across teams you do not control.
Advanced · production systems
44 sections · 5 parts + event studio
About 220 min reading
- Partitions and Keys
- Consumer and Share Groups
- Delivery Semantics
- Schema Evolution
Systems Engineering Track
AWS Architecture for Backend Engineers
Size an address space against a workload rather than a habit, walk an IAM decision to the layer that made it, place a service on the compute spectrum with its constraint named, and multiply a request path's availability instead of quoting one component's.
Advanced · production systems
38 sections · 4 parts + cloud studio
About 200 min reading
- VPC and Routing
- IAM Evaluation
- Compute Spectrum
- Availability Math
Systems Engineering Track
AWS Data, Messaging and Cost
Explain a throttle that happens with capacity to spare, choose a messaging service by what survives consumption, size a visibility timeout knowing it is only a mitigation, and read a bill as the document that describes your architecture.
Advanced · production systems
37 sections · 4 parts + data studio
About 185 min reading
- Managed Databases
- Hot Partitions
- Queues and Streams
- Cost as Design
Systems Engineering Track
Containers, Kubernetes and Infrastructure as Code
Order a build so a change rebuilds only what depends on it, tell Pending from throttled from OOMKilled by symptom, put a dependency check where it degrades rather than restarts, and justify every line of a plan's destroy list.
Advanced · production systems
41 sections · 3 parts + platform studio
About 205 min reading
- Image Layers
- Requests and Limits
- Probes
- Terraform State
Systems Engineering Track
Observability and Production Engineering
Say which statistic can see which failure and which cannot in principle, price a metric before adding a label, turn an SLO into an alert that fires on damage rather than a threshold, and follow one identifier from a burn alert to the line that explains it.
Advanced · production systems
35 sections · 4 parts + signals studio
About 180 min reading
- Percentiles
- Cardinality
- Error Budgets
- Tracing
Systems Engineering Track
Modern System Design
Estimate before you draw and name the assumption the answer turns on, resolve a skewed distribution with a measured threshold, choose a rate limiter by the burst it permits, and state the four-clause trade for every component in a design.
Advanced · production systems
38 sections · 4 parts + design studio
About 200 min reading
- Estimation
- Skew and Thresholds
- Rate Limiting
- Defending a Design
1 course
Workplace AI
Practical, verified AI workflows for analysts, operators, and business teams.
Learning Path
Progress without losing the system view.
Four core phases move from mental models to accountable architecture. Focused tracks then deepen the application, agent, infrastructure, or platform systems your role requires.
See the academy learning path- 01
Build the foundations
AI history, mathematics, machine learning, and deep learning mental models.
- 02
Engineer generative systems
LLMs, prompt systems, retrieval pipelines, evaluation, and model serving.
- 03
Design agentic workflows
Tools, memory, planning, MCP, orchestration, and multi-agent coordination.
- 04
Lead in production
Architecture, reliability, security, cost control, governance, and risk.
Engineering, Not Hype
Learn the decisions behind reliable AI systems.
Mental models first
Understand models, retrieval, agents, and evaluation as connected systems rather than isolated APIs.
Build to understand
Move from executable examples and labs to production failure modes, observability, and operational trade-offs.
Architect for reality
Design authority boundaries, security controls, cost limits, and governance before systems reach production.
Created from engineering practice
Technical education with accountable boundaries.
KnowledgeOS brings the disciplines used in serious software systems - explicit boundaries, measurable behavior, failure recovery, security, and operational evidence - into AI engineering education. The material is designed to help learners move beyond API familiarity and reason about complete production systems.
KnowledgeOS is an independent educational project. It is not affiliated with, sponsored by, reviewed by, or endorsed by Visa Inc.
Start with the full map
Build a durable AI engineering career, one decision at a time.
Compare the twelve-volume core with all focused specialization tracks, then choose the entry point and depth that match your current evidence and responsibilities.