AI Engineer Interview Preparation (Comprehensive)

A hands-on, 2026-ready curriculum for AI Engineer / ML Engineer interviews — 18 module notes, 38 runnable Python labs, and 9 portfolio projects covering classical ML through LLMs, RAG, agents, and production LLMOps.

This repository is a complete, hands-on tutorial to prepare for AI Engineer / ML Engineer interviews.

It is designed for:

• Beginners with little experience who want a clear path
• Engineers switching into AI/ML
• Candidates preparing for end-to-end AI Engineering interviews (ML + Deep Learning + LLMs + MLOps + System Design)

How to Use This Repo

• Follow the roadmap in docs/ROADMAP.md.
• For the full navigation index, use INDEX.md.
• Use docs/MODULES.md to jump into a module.
• If you have 3 months, follow docs/3_MONTH_PLAN.md.
• For FAANG-style preparation (coding + ML + system design), follow docs/3_MONTH_PLAN_FAANG.md.
• For each module, read the notes and then run the practice code in src/.
• Use interview/ for question banks, checklists, and mock interview practice.

Repository Map

• docs/
  • Guides and structured learning path
• interview/
  • Interview questions, cheat sheets, mock interview loops
• src/
  • Runnable code examples (from-scratch + PyTorch)
• projects/
  • Mini-projects you can build and discuss in interviews

Projects (recommended)

Start here once you finish the fundamentals in docs/.

python projects/01_churn_prediction/train.py      # classical ML end-to-end
python projects/02_rag_baseline/run.py            # minimal RAG
python projects/03_monitoring_drift/drift_check.py  # production drift
python projects/04_ranking_baseline/train.py      # learning to rank
python projects/05_time_series/baseline.py        # forecasting
python projects/06_mini_llm_eval/run.py           # LLM eval harness
python projects/07_agent_tool_use/run.py          # ReAct agent with tools
python projects/08_hybrid_rag/run.py              # BM25 + dense + rerank + RAGAS
python projects/09_llm_gateway/run.py             # guardrails + cache + budgets

Hands-on labs for modern AI Engineer interviews (2026)

Every script below is self-contained, runnable, and maps to a topic that 2026 AI Engineer interviews at FAANG and startups drill on.

Transformers, inference, and training internals

• Attention from scratch -- src/deep_learning/attention_from_scratch.py
• Transformer encoder block (PyTorch) -- src/deep_learning/transformer_encoder_torch.py
• Backprop from first principles -- src/deep_learning/autograd_micro.py
• Rotary Positional Embeddings (LLaMA default) -- src/deep_learning/rope_positional.py
• KV cache -- src/deep_learning/kv_cache.py
• Speculative decoding -- src/deep_learning/speculative_decoding.py
• LoRA -- src/deep_learning/lora_from_scratch.py
• DPO (preference learning) -- src/deep_learning/dpo_from_scratch.py
• Knowledge distillation -- src/deep_learning/knowledge_distillation.py
• CLIP-style contrastive dual encoder -- src/deep_learning/clip_dual_encoder.py
• INT8 quantization -- src/advanced/quantization_int8.py

LLM application layer

• ReAct agent loop -- src/llm/agent_react_loop.py
• Prompt patterns (few-shot, CoT, self-consistency) -- src/llm/prompt_patterns.py
• Structured output with self-healing retry -- src/llm/structured_output.py
• Streaming tokens, stop sequences, incremental JSON -- src/llm/streaming_generator.py
• Guardrails (PII, prompt injection, output policy) -- src/llm/guardrails.py
• Token economics and cost budgeting -- src/llm/token_economics.py
• Span-based tracing for RAG and agents -- src/llm/observability_tracing.py

Production RAG

• Hybrid BM25 + dense + RRF -- src/rag/hybrid_search_bm25.py
• Cross-encoder reranking -- src/rag/cross_encoder_rerank.py
• Semantic cache -- src/rag/semantic_cache.py
• RAGAS-style eval (faithfulness / relevancy / precision / recall) -- src/rag/ragas_style_eval.py
• Chunking strategies -- src/rag/chunking_strategies.py
• Dense embeddings + k-NN -- src/rag/embeddings_knn_search.py

Foundations

• Tokenization (BPE) -- src/advanced/bpe_tokenizer.py
• LLM decoding strategies -- src/advanced/llm_sampling.py
• Class imbalance & threshold tuning -- src/advanced/class_imbalance.py
• A/B test statistics -- src/advanced/ab_test_significance.py

Matching module notes

• docs/modules/12_transformer_internals.md
• docs/modules/13_llm_inference_and_finetuning.md
• docs/modules/14_llm_agents_and_tool_use.md
• docs/modules/15_production_rag_patterns.md
• docs/modules/16_llm_observability_cost_safety.md
• docs/modules/17_modern_training_and_alignment.md
• docs/modules/18_ai_engineer_2026_capabilities.md -- role map for FAANG, LLM labs, startups, and enterprise teams.

Interview question banks

• interview/llm_agents_and_production_questions.md -- 24 Qs on agents, production RAG, guardrails, cost, alignment.
• interview/genai_system_design_cases.md -- 10 worked case studies.

Quickstart

1) Create a virtual environment

python -m venv .venv

2) Activate it

Windows (PowerShell):

.venv\Scripts\Activate.ps1

macOS / Linux:

source .venv/bin/activate

3) Install dependencies

pip install -r requirements.txt

4) Run an example

python src/basics/linear_regression_numpy.py

Other examples:

python src/basics/metrics_from_scratch.py
python src/sql/sql_practice_sqlite.py
python src/rag/tfidf_rag_demo.py

What You’ll Learn

• ML fundamentals (bias/variance, regularization, metrics, CV)
• Data work (pandas, feature engineering, leakage, imbalance)
• Deep Learning (PyTorch training loops, debugging, optimization)
• LLM fundamentals (tokenization, transformers, finetuning concepts)
• RAG (chunking, retrieval, reranking, evaluation)
• MLOps/LLMOps (experiment tracking concepts, CI checks, monitoring)
• ML system design (requirements → data → modeling → serving → monitoring)

Validate the Repo

Run the full smoke test (checks imports and executes every key script end-to-end):

python validate.py

Contributing

Issues and PRs welcome. If you add a script under src/ or a new project, please also add it to validate.py and INDEX.md so the smoke test stays complete.

License

MIT — see LICENSE.