> ## Documentation Index > Fetch the complete documentation index at: https://resources.devweekends.com/llms.txt > Use this file to discover all available pages before exploring further. # AI Engineering > Build production-ready AI applications with Python, LLMs, agents, and modern AI infrastructure ## The \$500K+ Skill in 2025 Here is the uncomfortable truth about the AI job market: there are hundreds of thousands of people who can prompt ChatGPT, but very few who can build a production AI system that handles 10,000 concurrent users, gracefully degrades when the API goes down, costs less than \$500/month to run, and does not hallucinate confidently wrong answers to customers. That gap between "I used an LLM" and "I shipped an LLM product" is where the money is. AI Engineers are the highest-paid developers in tech right now. According to [Levels.fyi](https://levels.fyi), senior AI Engineers at top companies earn **$400K-$700K+ TC**. Why? Because companies are desperate for people who can actually **ship** AI products -- not just experiment with ChatGPT. **December 2025 Update**: This course now covers GPT-4.5, Claude 3.5 Opus, Gemini 2.0 Flash, the OpenAI Responses API, and the latest agentic patterns including computer use and MCP integrations. This is not another "prompt engineering" course. You will build **real systems** -- the kind that survive contact with production traffic, impatient users, and unreliable APIs: * A production RAG pipeline that handles 100K+ documents with hybrid search * Multi-agent systems with LangGraph that automate complex workflows * MCP servers that connect AI to databases, APIs, and external tools * AI applications with proper error handling, caching, and observability **What makes this different?** Every module includes production code you can deploy. No toy examples. No "hello world" chatbots. ### What Companies Are Building Right Now | Company Type | AI Applications | Your Skills | Typical Comp Range | | --------------------- | ------------------------------------------------------- | --------------------------------------------- | ---------------------------- | | **Startups (Seed-B)** | AI copilots, document automation, AI-native products | RAG, Agents, APIs, full-stack AI | $150K-$300K + equity | | **Enterprise** | Knowledge bases, workflow automation, internal tools | Vector DBs, Multi-agent, compliance, security | $200K-$400K TC | | **Dev Tools** | Code assistants, MCP integrations, developer experience | Tool use, LangGraph, SDK design | $250K-$500K TC | | **SaaS** | AI features, smart search, AI-powered analytics | Embeddings, Caching, multi-tenancy | $180K-$350K TC | | **Consulting/Agency** | Custom AI solutions across industries | Breadth across all modules | $150K-$400K (billable rates) | ## Prerequisites (Crash Courses Included) New to Python or backend development? We've got you covered: Types, async, classes, decorators—everything for AI work Build production APIs: routing, streaming, auth, deployment PostgreSQL, SQLAlchemy, pgvector, async patterns ## What You'll Build Production RAG system with hybrid search, re-ranking, and citations. Handles PDFs, docs, and web pages at scale. Agent that reviews PRs, suggests fixes, and explains changes. Uses function calling, structured outputs, and tool use. Multi-agent system using LangGraph that researches topics, synthesizes information, and writes comprehensive reports. Build an MCP server that gives AI models access to your PostgreSQL database with read/write capabilities. Full-stack AI document assistant with multi-tenancy, usage tracking, and real-time streaming—your portfolio piece. Agent that can control a browser/desktop to automate tasks using Anthropic's computer use capabilities. ## Course Modules How LLMs actually work. Tokenization, embeddings, attention, inference costs. System prompts, few-shot learning, chain-of-thought, and prompt optimization. Master function calling, structured outputs, streaming, vision, and cost optimization. pgvector, Pinecone, hybrid search, chunking strategies, and indexing. Build RAG that actually works. Query expansion, re-ranking, evaluation. When and how to fine-tune. OpenAI fine-tuning, LoRA, QLoRA patterns. ReAct agents, tool use, memory systems, and autonomous workflows. Chains, prompts, memory, tools, and production patterns with LangChain. Complex workflows, human-in-the-loop, parallel execution, state management. Build MCP servers to connect AI to databases, APIs, and tools. Multi-agent architectures, supervisor patterns, and production design. LLM-as-Judge, eval pipelines, CI/CD for AI, and monitoring. Caching, rate limiting, model routing, and production infrastructure. Build DocuMind AI—a complete AI SaaS from scratch. ## Learning Path **For those new to Python/Backend** * Python crash course: async, types, classes * FastAPI: APIs, streaming, dependency injection * Databases: PostgreSQL, SQLAlchemy, migrations **Skip if**: You already know Python and have built APIs **Goal**: Understand LLMs deeply, not superficially. You should be able to explain *why* a prompt works, not just *that* it works. * How transformers and attention work (the intuition, not the math) * Tokenization, context windows, and costs (why "strawberry" tokenization matters) * Embeddings and semantic similarity (the backbone of search and RAG) * Prompt engineering that actually works (tested techniques, not Twitter tips) **Project**: Build a cost-aware chat application **Goal**: Master the OpenAI API beyond basics. * Streaming responses for real-time UX * Function calling for structured actions * Structured outputs with Pydantic * Vision and multimodal inputs **Project**: AI Code Reviewer that analyzes GitHub PRs **Goal**: Build RAG systems that don't suck. Most RAG tutorials produce systems that work on demos and fail on real documents -- you will learn why and how to fix it. * Chunking strategies that preserve context (the #1 RAG mistake is bad chunking) * Hybrid search (semantic + keyword) -- because neither alone is sufficient * Re-ranking for precision (retrieve broadly, then filter aggressively) * Evaluation and continuous improvement (if you cannot measure it, you cannot improve it) **Project**: Production document Q\&A system **Goal**: Deploy AI systems that scale. * Agent architectures and patterns * LangGraph for complex workflows * MCP for tool integration * Caching, rate limiting, observability **Project**: Multi-agent research assistant ## Prerequisites Classes, async/await, type hints, virtual environments. No ML experience needed. SELECT, JOIN, indexes. We'll use PostgreSQL with pgvector. HTTP methods, JSON, headers. We'll build FastAPI services. Navigate directories, run scripts, use git. Docker is a plus. ## Tech Stack (2025 Edition) | Category | What You'll Use | | ------------------ | ------------------------------------------------------------------------------------- | | **Languages** | Python 3.12+, TypeScript (optional), SQL | | **LLM Providers** | OpenAI (GPT-4.5, GPT-4o), Anthropic (Claude 3.5), Google (Gemini 2.0), Ollama (local) | | **Frameworks** | LangChain 0.3+, LangGraph, FastAPI, Pydantic v2 | | **Vector DB** | pgvector (PostgreSQL 16+), Pinecone, Chroma, Qdrant | | **Protocols** | Model Context Protocol (MCP), OpenAI Responses API | | **Infrastructure** | Docker, Redis, PostgreSQL, Supabase | | **Observability** | LangSmith, Langfuse, OpenTelemetry | ## Who Is This For? You can code but haven't built AI systems. You want to add AI features to products or transition into AI engineering. This course takes you from "I've used ChatGPT" to "I ship AI products." You build APIs and services. You want to add LLM capabilities—chatbots, document search, automation. You'll learn to integrate AI while maintaining the reliability you're used to. You know ML but struggle with production deployment. Notebooks are great for experimentation, but you want to build real applications. This course bridges the gap. You need to build AI features fast. You can't afford to hire a team of specialists. This course gives you the skills to prototype and ship AI products yourself. ## What You'll Walk Away With Production-ready projects you can demo to employers or use in your products. Templates and patterns you can copy into any project. Know *why* things work, not just how to copy-paste. ## Module Dependencies Not every module needs to be completed in order. Here is a map of what depends on what, so you can skip ahead if you already have the prerequisites: | Module | Hard Prerequisites | Soft Prerequisites (helpful but not required) | | -------------------- | ----------------------------- | --------------------------------------------- | | LLM Fundamentals | None | Python basics | | Prompt Engineering | LLM Fundamentals | None | | OpenAI API | Python basics | Prompt Engineering | | Vector Databases | Python basics, SQL | LLM Fundamentals (embeddings) | | RAG Systems | Vector Databases, OpenAI API | Prompt Engineering | | Fine-Tuning | LLM Fundamentals, OpenAI API | RAG (to know when NOT to fine-tune) | | AI Agents | OpenAI API (function calling) | Prompt Engineering | | LangChain | OpenAI API | RAG, Agents | | LangGraph | LangChain, Agents | RAG | | MCP Protocol | OpenAI API (tool use) | LangGraph | | Evaluation & Testing | RAG or Agents | Prompt Engineering | | Deployment & Scaling | Any 2 building modules | All building modules | | Capstone | All above | All above | ## Start Here **Recommended Path**: Start with LLM Fundamentals, then work through each section in order. Each module builds on the previous one. If you are experienced with LLMs, skip to OpenAI API or RAG Systems. Understand how large language models work before building with them