AI Engineer

Goal: Build the engineering foundation: Python for AI systems, tensors/autograd, training engineering and tokenization.

1. Day01

   Map an AI application into data, model, inference, evaluation and production components.

2. Day02

   Write Python suitable for AI workflows, including data handling, functions, modules and practical pipeline code.

3. Day03

   Use tensors and autograd to understand and implement basic differentiable computations.

4. Day04

   Explain a training loop and identify the roles of data, batches, loss, optimizer and checkpoints.

5. Day05

   Inspect tokenization and explain how text becomes model-ready token IDs.

Goal: Understand the core Transformer mechanisms and the engineering implications of inference.

1. Day06

   Explain learned representations and inspect embedding behavior for semantic similarity.

2. Day07

   Implement or trace self-attention and explain how token relationships are computed.

3. Day08

   Explain why multiple attention heads can capture different relationships in the same sequence.

4. Day09

   Trace the major components of a Transformer block and explain their purpose.

5. Day10

   Explain inference bottlenecks and the practical trade-offs affecting latency and throughput.

Goal: Engineer LLM applications using deliberate model choice, structured generation and scalable prompting.

1. Day11

   Select models using task quality, latency, context, reliability and cost requirements.

2. Day12

   Design reliable structured outputs and validate model responses before application use.

3. Day13

   Create reusable prompt patterns, templates and versioned prompt workflows.

4. Day14

   Define an engineering problem, architecture, success criteria and test plan for Mini Project 1.

5. Day15

   Build the first engineering-grade version of Mini Project 1 with structured outputs and tests.

Goal: Build an advanced RAG pipeline from ingestion through retrieval and reranking.

1. Day16

   Improve the project using test results, failure analysis and engineering changes.

2. Day17

   Design the end-to-end RAG architecture and identify its quality and latency bottlenecks.

3. Day18

   Build a document ingestion pipeline that cleans, chunks and prepares content for retrieval.

4. Day19

   Implement retrieval and tune it against representative queries.

5. Day20

   Add reranking and explain how it can improve the relevance of retrieved context.

Goal: Improve retrieval reliability and decide when to use prompting, RAG or fine-tuning.

1. Day21

   Transform difficult user queries into retrieval-friendly queries and evaluate the improvement.

2. Day22

   Measure retrieval and answer quality with a repeatable evaluation approach.

3. Day23

   Identify and mitigate common RAG failure modes such as missing context, noisy context and unsupported answers.

4. Day24

   Choose an adaptation approach based on knowledge freshness, behavior, data and engineering constraints.

5. Day25

   Explain the fine-tuning workflow and identify when it is appropriate.

Goal: Understand practical model adaptation and build increasingly capable tool-using agents.

1. Day26

   Explain parameter-efficient fine-tuning and build or trace a simple LoRA/PEFT workflow.

2. Day27

   Evaluate whether model adaptation actually improved the target behavior.

3. Day28

   Engineer a tool-using agent with clear tool schemas and controlled execution.

4. Day29

   Design memory for an agent and distinguish short-term state from longer-lived knowledge.

5. Day30

   Explain how standardized tool interfaces can connect agents to broader tool ecosystems.

Goal: Make agents measurable and safer while extending them to multimodal and voice workflows.

1. Day31

   Create tests for agent trajectories, tool use, task success and failure cases.

2. Day32

   Engineer a workflow that combines text with image or other modalities.

3. Day33

   Build or trace a voice workflow and reason about latency, turn-taking and failure handling.

4. Day34

   Identify prompt injection, data exposure and tool misuse risks and apply engineering controls.

5. Day35

   Instrument an AI workflow so failures and quality regressions can be investigated.

Goal: Move from an engineered prototype toward a production AI system with explicit architecture, cost and reliability decisions.

1. Day36

   Estimate major AI costs and identify changes that can reduce latency or spend.

2. Day37

   Design a production architecture with APIs, model services, data, retrieval, evaluation and operational controls.

3. Day38

   Defend the capstone architecture and justify key technical trade-offs.

4. Day39

   Implement and integrate the capstone system against its defined success criteria.

5. Day40

   Review failures and decisions, document the system and present the engineering outcome.