8 Steps to Becoming an AI Engineer: DIY Bootcamp with Kirill Eremenko

AI Engineering Bootcamp Overview

• 🚀 The SuperDataScience AI Engineering Bootcamp is an 8-week program designed to take participants from an intermediate to an advanced level.
• 💡 The goal is to equip individuals with the tools and approaches necessary to become effective AI engineers, focusing on solving business problems.
• 🎯 Prerequisites include strong Python skills, familiarity with PyTorch and scikit-learn, understanding of LLM API calls, and some cloud experience (AWS recommended).

Weeks 1-4: AI Science and Prototyping

Week 1: Mindset Shift

• 🧠 Focuses on understanding the problem before building a solution, assessing if AI or LLMs are truly needed.
• 💡 Participants compare different LLMs and explore reasoning vs. chat models to understand their use cases.
• ⚠️ Pro Tip: Define the business goal and success metric before building any AI solution.

Week 2: Behavior Design

• 🛠️ Designing how an LLM behaves through prompt templates, system prompts, and understanding response formats (e.g., JSON).
• ✈️ Participants build a flight assistant application using Gradio to create a visual interface for an LLM.
• 🔑 Pro Tip: The system prompt is always guaranteed to be passed to the LLM, making it crucial for controlling behavior, especially when context windows are large.

Week 3: RAG Foundations

• 📚 Explores Retrieval Augmented Generation (RAG) as a shift from fine-tuning to inference-time customization for LLMs.
• 🔗 Participants build a RAG pipeline from scratch, learning about chunking, embeddings, vector databases, and retrieval chains.
• 📈 Pro Tip: Smart chunking, considering overlap and semantic clarity, significantly improves RAG performance.

Week 4: Agentic AI

• 🤖 Introduces AI agents as an LLM brain with access to memory, RAG, and tools, enabling them to perform actions.
• 🔗 Agents interact with tools via text-based prompts, with the system code facilitating the actual tool execution.
• 🎓 Pro Tip: Route tool calls to the correct tools via code rather than relying solely on the LLM's decision-making.

Weeks 5-8: AI Deployment and Production

Week 5: Production Readiness

• ☁️ Focuses on transitioning from prototypes to production-ready applications, including setting up AWS CLI, infrastructure as code, and Docker.
• 🔄 Pro Tip: Implement a caching layer around LLM calls to save costs, speed up responses, and stabilize output.

Week 6: Memory and Security

• 🧠 Adding long-term memory to LLMs to enhance agent capabilities and making architectural decisions about memory databases.
• 🔒 Participants attempt to trick their deployed flight assistant agent into issuing refunds for non-refundable tickets, highlighting security vulnerabilities.
• ⚠️ Pro Tip: Don't solely trust prompts for security constraints; log and verify tool calls before executing sensitive actions.

Week 7: Knowledge RAG in Production

• 🌐 Discusses implementing RAG in production environments, contrasting it with proof-of-concept scenarios.
• 🔍 Pro Tip: Design queries for RAG systems to account for potential semantic mismatches between user questions and document content.

Week 8: Capstone Project

• 🎓 Participants finalize their AI agent (e.g., a digital twin) as a functioning product in a production AWS environment with CI/CD workflows.
• 🚀 The theme is treating AI agents as evolving products, not just projects, by setting up systems to identify and address knowledge gaps.
• ✅ Pro Tip: Continuously evolve your AI agent as a product by incorporating feedback and updating its knowledge base.