PERSONAL AI LEARNING PORTAL

Build a deeper
understanding.

A focused path from modern deep-learning internals to reliable, production-grade AI systems.

Start the path ↓Browse all resources ↗

08learning phases

16weeks of practice

28curated resources

01capstone system

THE CURRICULUM

Learn in layers.

Each phase pairs a small set of high-quality resources with implementation work, evidence, and an exit criterion.

01Weeks 1–3

Build the engine

Modern deep learning foundations

Go beneath the framework layer: gradients, optimization, training systems, and reproducible experiments.

02Weeks 4–6

Understand the architecture

Transformers & language models

Build the components behind modern language models and connect architecture to compute and memory.

03Weeks 7–9

Ship useful systems

Applied LLM engineering

Build a grounded assistant with retrieval, citations, streaming responses, and carefully scoped tools.

04Weeks 10–11

Measure what matters

Evaluation & reliability

Turn demos into dependable products with datasets, error analysis, metrics, and regression tests.

THE LIBRARY

Start with the
right signal.

⌕

▰CourseAndrej KarpathyNeural Networks: Zero to HeroImplement backpropagation, neural nets, and GPT from scratch.Modern deep learning foundations ↗▰CourseStanfordStanford CS336: Language Modeling from ScratchA rigorous, implementation-heavy path from tokenizer to deployment.Modern deep learning foundations ↗↗DocumentationPyTorchAutomatic mixed precisionTrain faster and with a smaller memory footprint.Modern deep learning foundations ↗▰CourseStanfordCS336 lectures & assignmentsTokenization, attention, scaling, data, and distributed training.Transformers & language models ↗▰CourseHugging FaceHugging Face LearnPractical Transformers, datasets, tokenizers, and Accelerate.Transformers & language models ↗◎PaperVaswani et al.Attention Is All You NeedThe original Transformer architecture paper.Transformers & language models ↗◎PaperDao et al.FlashAttentionMemory-efficient exact attention and GPU-aware thinking.Transformers & language models ↗▰CourseDeepLearning.AIRetrieval-Augmented GenerationDesign, build, evaluate, and deploy production-ready RAG systems.Applied LLM engineering ↗▰CourseFull Stack Deep LearningFull Stack LLM BootcampLLM foundations, product design, augmented language models, and LLMOps.Applied LLM engineering ↗▰CourseHugging FaceAgents CourseTools, agent loops, agentic RAG, observability, and evaluation.Applied LLM engineering ↗▰CourseDeepLearning.AIAdvanced Retrieval for AIQuery improvement, weak retrieval diagnosis, and feedback.Applied LLM engineering ↗↗GuideHamel Husain & Shreya ShankarLLM Evals: Everything You Need to KnowA practical guide to traces, error analysis, judges, and evaluation lifecycles.Evaluation & reliability ↗▰CourseHamel Husain & Shreya ShankarAI Evals courseAnalyze → Measure → Improve for real AI products.Evaluation & reliability ↗▰CourseDeepLearning.AI + W&BEvaluating and Debugging Generative AITracking, tracing, versioning, and debugging generative systems.Evaluation & reliability ↗◎PaperEs et al.RAGASReference-free evaluation for retrieval-augmented generation.Evaluation & reliability ↗↗DocumentationHugging FacePEFT documentationEfficiently adapt large pretrained models with methods such as LoRA.Fine-tuning & adaptation ↗↗DocumentationHugging FaceTRL documentationSFT, DPO, GRPO, reward modeling, and post-training workflows.Fine-tuning & adaptation ↗◎PaperHu et al.LoRALow-rank adaptation for efficient fine-tuning.Fine-tuning & adaptation ↗◎PaperDettmers et al.QLoRAMemory-efficient quantized fine-tuning.Fine-tuning & adaptation ↗▰CourseMade With MLMLOps CourseDesign, train, evaluate, serve, test, and monitor production ML.Production AI systems ↗▰CourseFSDLFull Stack Deep Learning 2022Experiment management, deployment, monitoring, and continual learning.Production AI systems ↗↗DocumentationvLLMvLLM documentationHigh-throughput serving, batching, caching, quantization, and parallelism.Production AI systems ↗↗GuideChip HuyenMachine Learning Systems DesignFree notes on data, training, serving, and system trade-offs.Production AI systems ↗↗GuideOWASPOWASP GenAI Security ProjectCurrent risks and mitigations for LLM and GenAI applications.Security & responsible deployment ↗↗GuideOWASPPrompt injection preventionPractical defenses for direct and indirect prompt injection.Security & responsible deployment ↗↗GuideOWASPOWASP MCP Top 10Security risks for tool-connected and MCP-enabled systems.Security & responsible deployment ↗⌁BuildYour projectRepository-level engineering assistantSearch code and docs, cite answers, run scoped tools, and report evidence.Capstone: engineering assistant ↗↗ProjectHugging FaceHugging Face Agents final projectUse benchmark-driven agent development as a model for your final project.Capstone: engineering assistant ↗

THE METHOD

Understand it.
Build it. Test it.

01Learn the mechanism

Read, watch, and explain the idea in your own words.

02Implement the minimum

Build a small version before reaching for an abstraction.

03Measure the result

Use evidence, failure analysis, and a written conclusion.

THE NORTH STAR

One system.
Many lessons.

Carry one repository-level engineering assistant through the path. Let every phase make the same product more capable, reliable, and honest.

View capstone plan ↗