Skip to content

Latest commit

 

History

27 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI Engineering from First Principles

Build Reliable Intelligent Systems That Improve Real Work

Take a messy real-world workflow, decide whether AI is useful, design a reliable intelligent system, build it, evaluate it, deploy it responsibly, and improve it with feedback.

Reader Promise

After finishing this course, a learner should be able to:

  • analyze a real workflow;
  • decide when not to use AI;
  • design system boundaries;
  • build AI-ready software foundations;
  • organize data, documents, and state;
  • use models as bounded components;
  • build grounded retrieval systems;
  • work with multimodal documents and human correction;
  • connect AI to tools, workflows, approvals, and agents;
  • evaluate quality, cost, latency, traces, and failure modes;
  • deploy, roll back, and improve AI systems responsibly;
  • reason about security, privacy, governance, and economics;
  • ship proof of useful capability through portfolio-grade projects.

Book Shape

  • 46 chapters across 9 ordered parts.
  • 9 guided build projects, one after each major capability.
  • 1 capstone that proves end-to-end AI engineering capability.
  • Chapters teach the mental model, mechanics, examples, mistakes, design decisions, build artifact, recall questions, summary, and next-step preview.
  • Projects are build-ready specifications, not exercises. Each project is scoped only to concepts already taught.
  • The capstone asks the learner to choose one meaningful workflow, build a narrow system, evaluate it, review risk, and make a product-real launch decision.

Each chapter ends with a proof artifact. No chapter ends with only "understanding."

Projects

The guided projects are the portfolio path through the course:

  1. Problem-first workflow analysis.
  2. AI-ready app skeleton.
  3. Knowledge base search engine.
  4. Structured model extraction.
  5. Grounded RAG assistant.
  6. Multimodal document analyst.
  7. Tool and agent workflow assistant.
  8. Production AI control center.
  9. Risk, governance, and economics review.

Start here: Project Guide

License & Contributions

The project is open source under Apache-2.0. Quainy names, logos, and brand assets are not licensed for reuse except for reasonable attribution as described in the contributing guide and notice.

Table of Contents

Part 1: Problem-First AI Engineering

Capability: decide what to build and why.

  1. AI Engineering Is System Design
  2. When Not to Use AI
  3. From Messy Workflow to System Boundary
  4. Metrics, Baselines, and Meaningful Impact

Project: Support Workflow Intelligence Console

Part 2: Software Foundations for AI Systems

Capability: understand where AI components live inside real software.

  1. AI Application Architecture
  2. APIs, Services, Workers, and Queues
  3. Auth, Permissions, Files, and Storage
  4. Async Jobs, Background Processing, and Webhooks

Project: AI App Skeleton

Part 3: Data and Knowledge Foundations

Capability: turn messy knowledge into usable system context.

  1. Data, Documents, and State
  2. Data Ingestion, Parsing, and Cleaning
  3. Schemas, Contracts, and Structured Outputs
  4. Data Quality, Lineage, and Versioning
  5. Search, Ranking, and Information Architecture

Project: Knowledge Base Search Engine

Part 4: Models as Components

Capability: use models with understanding, not worship.

  1. What Language Models Actually Do
  2. Tokens, Context, Attention, and Generation
  3. Prompting as Interface Design
  4. Model Selection, Routing, and Fallbacks
  5. Fine-Tuning, Adaptation, and When Not to Tune
  6. Model Failure Modes and Error Analysis

Project: Structured Intake Extractor

Part 5: Retrieval and Context Systems

Capability: build systems that answer from trusted knowledge.

  1. Embeddings and Semantic Retrieval
  2. Chunking, Indexing, and Retrieval Pipelines
  3. Retrieval-Augmented Generation
  4. Context Engineering and Grounding
  5. Evaluating Grounded AI Systems
  6. From Assistant Demo to Knowledge Product

Project: Grounded Research Assistant

Part 6: Multimodal and Human Interfaces

Capability: handle documents, images, audio, realtime inputs, and human correction loops.

  1. Document AI: OCR, Tables, Forms, and PDFs
  2. Vision, Image, and Multimodal Inputs
  3. Audio, Speech, and Realtime AI
  4. AI UX, Trust, Feedback, and Human Correction

Project: Multimodal Document Analyst

Part 7: Tools, Workflows, and Agents

Capability: build AI systems that take bounded action.

  1. Tool Use and API Actions
  2. Connectors, MCP, and Tool Registries
  3. Workflow Orchestration
  4. Memory, State, and Human Approval
  5. Computer Use and Browser Automation
  6. Agents Without Hype
  7. Agent Evaluation and Trace Review

Project: Operations Action Assistant

Part 8: Reliable Production AI

Capability: measure, observe, deploy, and improve AI systems without guessing.

  1. Evaluation as Product Infrastructure
  2. Observability, Traces, Cost, and Latency
  3. Caching, Streaming, Batching, and Inference Optimization
  4. LLMOps: Versioning Prompts, Models, Evals, Data, and Indexes
  5. Deployment, Rollbacks, Feedback, and Continuous Improvement

Project: Production AI Control Center

Part 9: Security, Governance, and Product Reality

Capability: make AI systems safe, governed, economically justified, and worth keeping.

  1. AI Security Threat Model
  2. Prompt Injection, Retrieval Poisoning, and Tool Abuse
  3. Privacy, PII, Data Retention, and Access Control
  4. Responsible AI, Red Teaming, and Risk Management
  5. AI Product Economics: Cost, ROI, Build-vs-Buy

Project: AI Risk and Governance Review

Final Capstone

Final proof: choose one real workflow, build one narrow intelligent system, evaluate it, analyze failure, document risk, estimate economics, and show deployment readiness.

Build One Real Intelligent System

About

Take a messy real-world workflow, decide whether AI is useful, design a reliable intelligent system, build it, evaluate it, deploy it responsibly, and improve it with feedback.

Topics

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors