Neuriwe
Curriculum
Five tracks, fifteen courses.
90 modules across 15 courses, each taught for Studio and Labs learners.
How the curriculum works
One module a week. Two outcomes for every module.
Each course runs six modules, one per week, plus a final assessment. Every module has a Studio (Builder) outcome and a Labs (Engineer) outcome, and both streams meet in the course project review.
Module content reflects the curriculum blueprint and is reviewed every quarter.
| Track | Courses | Modules |
|---|---|---|
| Track 1: Foundations and the History of Compute (1950s-2011) | 101, 102, 103 | 18 |
| Track 2: The Deep Learning Revolution (2012-2021) | 201, 202, 203 | 18 |
| Track 3: The Generative Era (2022-2025) | 301, 302, 303 | 18 |
| Track 4: The Agentic Era (2026) | 401, 402, 403 | 18 |
| Track 5: Future Horizons, AGI and Physical Reality (2027 and beyond) | 501, 502, 503 | 18 |
Track 1
Foundations and the History of Compute (1950s-2011)
Objective: Understand core algorithmic logic, rule-based systems, and the mathematics beneath modern AI. Level: Foundations.
Course 101: Symbolic AI, Knowledge Engineering and the Logic Framework
- Summary
- The logical foundation of AI before data-driven learning, and the limits that caused two AI winters.
- Prerequisites
- None
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- AI history and risk brief for your industry
- Labs project
- Rule-based expert system with forward and backward chaining
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 The Dartmouth Vision and Turing Tests | Dartmouth 1956, Turing test, symbolic processing | Explain what counts as AI and where claims overreach | Implement a Turing-style dialogue tester |
| M2 Logic and Theorem Proving | Propositional and first-order logic, resolution | Read logical rules inside business policies | Code a resolution prover for small knowledge bases |
| M3 Knowledge Representation | Semantic networks, frames, ontologies | Map company knowledge as a simple ontology | Model a domain with frames and an ontology |
| M4 Expert Systems | Production rules, forward vs backward chaining, inference engines | Judge when a rules engine beats machine learning | Build an inference engine |
| M5 The First AI Winter (1974-1980) | Lighthill Report, combinatorial explosion, compute limits | Spot hype cycles in vendor claims | Measure combinatorial explosion in search |
| M6 The Second AI Winter (1987-1993) | Lisp machine collapse, expert system fragility | Build a hype-versus-evidence checklist | Stress-test brittle rules on edge cases |
Course 102: Mathematical Foundations for AI
- Summary
- The linear algebra, calculus, and probability needed to build neural models from scratch.
- Prerequisites
- High-school mathematics (Studio needs no calculus)
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Metrics and data interpretation memo
- Labs project
- Gradient descent and SVD implemented from scratch
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Linear Algebra Foundations | Vector spaces, matrix multiplication, transformations, basis changes | Read vectors and matrices as data tables | Implement matrix operations in NumPy |
| M2 Spectral Theory | Eigenvalues, eigenvectors, SVD | Interpret dimensionality-reduction outputs | Compute SVD and apply it to compression |
| M3 Multivariate Calculus | Partial derivatives, gradients, Jacobians, Hessians, chain rule | Understand why models improve in small steps | Derive and code gradients by hand |
| M4 Optimization Principles | Convexity, extrema, Lagrange multipliers, gradient descent | Judge trade-offs when tuning systems | Implement gradient descent variants |
| M5 Probability Spaces | Conditional probability, Bayes, random variables, distributions | Reason about uncertainty and base rates | Simulate distributions and Bayesian updates |
| M6 Statistical Foundations | MLE, expectation, variance, covariance matrices | Read variance and confidence in reports | Fit models by maximum likelihood |
Course 103: Classical Statistical Learning and Empirical Decision Models
- Summary
- Non-neural machine learning, parameter tuning, and validation.
- Prerequisites
- 101 and 102, or equivalent
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Vendor selection and ROI scoping for an ML use case
- Labs project
- End-to-end scikit-learn pipeline with tuning
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Regression Models | OLS, residual analysis, collinearity | Interpret regression outputs in business terms | Fit and diagnose regression models |
| M2 Regularization | Ridge, Lasso, ElasticNet | Understand overfitting risk in vendor demos | Tune L1 and L2 penalties |
| M3 Classification | Logistic regression, SVM, kernel tricks | Frame decisions as classification problems | Train SVMs with kernels |
| M4 Non-Parametric Models | k-NN, decision trees, ID3, C4.5, CART | Explain tree-based decisions to stakeholders | Code a decision tree from scratch |
| M5 Ensemble Frameworks | Bagging, random forests, AdaBoost, gradient boosting, XGBoost | Weigh accuracy gains against cost | Build and compare ensembles |
| M6 Validation | Overfitting, bias-variance, cross-validation, ROC/AUC | Demand proper evaluation from vendors | Run cross-validation and ROC analysis |
Track 2
The Deep Learning Revolution (2012-2021)
Objective: Move from explicit programming to statistical learning and neural architectures. Level: Practitioner.
Course 201: Artificial Neural Networks and Optimization
- Summary
- The internal mechanics of gradient flow in multi-layer networks.
- Prerequisites
- Track 1
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Brief translating a business goal into a model type
- Labs project
- MLP trained in PyTorch with AdamW
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Perceptrons and Activations | Sigmoid, tanh, ReLU, GELU | Understand neurons as weighted decisions | Implement perceptrons and activations |
| M2 Feedforward Networks | MLPs, weight matrices, bias vectors | Read a model diagram | Build an MLP in PyTorch |
| M3 Backpropagation | Computational graphs, forward pass, reverse-mode differentiation | See how errors flow back to improve a model | Implement backprop by hand and with autograd |
| M4 Gradient Challenges | Vanishing and exploding gradients, mitigation | Recognize why training stalls | Diagnose and fix gradient problems |
| M5 Advanced Optimizers | Momentum, AdaGrad, RMSprop, Adam, AdamW | Understand training time and cost drivers | Compare optimizers on one task |
| M6 Regularization | Dropout, weight decay, batch and layer normalization | Know why models need guardrails | Add regularization and measure generalization |
Course 202: Spatial Processing and Computer Vision
- Summary
- From convolution to modern deep vision backbones and generative vision.
- Prerequisites
- 201
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- AI-driven branding and video workflow with a media-rights checklist
- Labs project
- Train and evaluate a CNN or YOLO detector with OpenCV
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Image Processing Foundations | Convolutions, kernels, stride, padding, pooling | See how machines read images | Implement convolution and pooling |
| M2 Early Vision Pipelines | AlexNet 2012, VGG, Inception | Trace the 2012 breakthrough | Reproduce a small AlexNet-style model |
| M3 Residual Networks | Skip connections, identity mappings, degradation | Judge model-depth claims | Build and train a ResNet block |
| M4 Detection and Localization | R-CNN family, YOLO | Specify detection use cases | Train a YOLO detector |
| M5 Dense Prediction | Semantic segmentation, FCNs, U-Net | Plan visual quality-assurance workflows | Build a U-Net |
| M6 Generative Vision | Autoencoders, VAEs, GANs | Use Midjourney and Runway with rights in mind | Train a VAE and a simple GAN |
Course 203: Sequence Models and the Transformer Leap
- Summary
- From recurrent networks to attention and the 2017 Transformer.
- Prerequisites
- 201
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Enterprise prompt framework with hallucination controls
- Labs project
- Self-attention and a small Transformer implemented
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Recurrent Networks | RNNs, BPTT challenges | Understand sequence data | Train an RNN on text |
| M2 Gated Architectures | LSTM, GRU | Know why early language tools forgot context | Implement LSTM and GRU cells |
| M3 Sequence-to-Sequence | Encoder-decoder, teacher forcing, context vectors | Map translation and summarization workflows | Build a seq2seq translator |
| M4 The Attention Catalyst | Bahdanau and Luong attention | Understand why attention changed everything | Add attention to seq2seq |
| M5 The Transformer (2017) | Scaled dot-product and multi-head attention | Write multi-step prompts that exploit context | Code QKV self-attention |
| M6 Positional Encoding | Sinusoidal and learned encodings, feed-forward sublayers | Handle context limits and hallucination risk | Assemble a mini Transformer |
Track 3
The Generative Era (2022-2025)
Objective: Master foundation language models, alignment, and multimodal content creation. Level: Practitioner.
Course 301: Architectural Engineering of Foundational LLMs
- Summary
- How internet-scale autoregressive models are built, pre-trained, and managed.
- Prerequisites
- 203
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Model comparison and cost evaluation for a use case
- Labs project
- Small language model pre-trained with a BPE tokenizer
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Model Families | Decoder-only (GPT), encoder-only (BERT), encoder-decoder (T5) | Choose model types for tasks | Compare architectures on benchmarks |
| M2 Tokenization | BPE, WordPiece, SentencePiece | Estimate cost through tokens | Train a BPE tokenizer |
| M3 Attention Optimization | FlashAttention, MQA, GQA | Understand speed and cost drivers | Implement efficient attention |
| M4 Position and Context | RoPE, ALiBi, long context windows | Plan around context limits | Extend context with RoPE |
| M5 Data Curation | Scraping at scale, MinHash/LSH dedup, quality filtering | Assess data risk in vendors | Build a dedup and filter pipeline |
| M6 Pre-training Engineering | Masked vs causal modeling, Chinchilla scaling laws, loss spikes | Budget compute realistically | Run a small pre-training experiment |
Course 302: Post-Training, Alignment and Model Behavior
- Summary
- How base models are aligned into steerable assistants.
- Prerequisites
- 301
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- No-code fine-tune and QA plan for a company assistant
- Labs project
- LoRA fine-tune with DPO and quantization
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Instruction Tuning | SFT, dataset curation | Design instruction data for a company assistant | Run supervised fine-tuning |
| M2 Human Feedback | RLHF, reward models | Plan human review loops | Train a reward model |
| M3 Preference Optimization | PPO vs DPO | Understand alignment trade-offs | Implement DPO |
| M4 Parameter-Efficient Tuning | LoRA, QLoRA, prefix and prompt tuning | Fine-tune open models with no-code interfaces | Fine-tune with LoRA and QLoRA |
| M5 Quantization | AWQ, GPTQ, GGUF, FP8/FP4 | Weigh cost against quality | Quantize for target hardware |
| M6 Safety Boundaries | Jailbreaks, red teaming, prompt injection | Write a QA and safety framework | Red-team and patch a model |
Course 303: Multimodal Systems and Diffusion Mechanics
- Summary
- Cross-modal projections and the physics of visual generation.
- Prerequisites
- 202 and 301
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Automated creative studio pipeline for ads
- Labs project
- Latent diffusion sampler or small vision-language demo
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Vision-Language Bridging | CLIP, contrastive learning, shared embeddings | Understand text-to-image alignment | Train a CLIP-style model |
| M2 Vision-Language Models | Projection layers, Perceiver resamplers, Flamingo, LLaVA | Plan multimodal assistants | Build a small VLM |
| M3 Diffusion Fundamentals | Forward and reverse diffusion, SDEs, score matching | Grasp how images emerge from noise | Implement denoising score matching |
| M4 Latent Diffusion | Stable Diffusion, VAE latents, U-Net cross-attention | Prompt for hyper-realistic media | Build a latent diffusion sampler |
| M5 Text-to-Video | Spatio-temporal Transformers, DiT | Produce cinematic ads with Runway or Sora | Prototype a DiT video block |
| M6 Native Multimodality | Unified text, audio, vision models | Design an end-to-end creative studio | Integrate multimodal inputs in one model |
Track 4
The Agentic Era (2026)
Objective: Move beyond static prompts to autonomous, multi-step execution and tool use. Level: Advanced.
Course 401: Advanced Reasoning and Search-Based Generation
- Summary
- Models that think before they answer, and enterprise retrieval.
- Prerequisites
- 301
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Enterprise knowledge base with accuracy checks
- Labs project
- GraphRAG system with evaluation
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Reasoning Topologies | Chain, tree, and graph of thought | Structure complex questions for AI | Implement CoT and ToT prompting |
| M2 Test-Time Compute | MCTS inside generation | Understand why slower can be smarter | Add MCTS search to generation |
| M3 Self-Correction | Critique loops, verification, self-reflection | Add review steps to workflows | Build self-correction loops |
| M4 Enterprise RAG | Retrieval-augmented generation, vector DBs, hierarchical indexing | Structure knowledge-management databases | Build RAG with a vector database |
| M5 Advanced Retrieval | Graph RAG, temporal chunking, multi-vector retrieval | Evaluate search accuracy and data safety | Build GraphRAG and optimize indexes |
| M6 Flagship Reasoning Models | Architecture patterns of leading reasoning models | Select models by task and cost | Benchmark reasoning models |
Course 402: Autonomous Agent Design and Multi-Agent Execution
- Summary
- Independent computational entities that execute complex tasks.
- Prerequisites
- 401
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Multi-agent sales and marketing automation in Make or Flowise
- Labs project
- Stateful multi-agent system in LangGraph with approval gates
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Agent Anatomy | ReAct, short-term and long-term memory | Define what an agent should and should not do | Implement ReAct with memory |
| M2 Tool Access | Function calling, code interpreters, sandboxing | Connect agents to business apps | Build tool-calling pipelines |
| M3 Multi-Agent Topologies | Boss-worker, peer-to-peer, orchestrators | Design cross-app workflows | Implement orchestrated agent teams |
| M4 Execution Frameworks | LangGraph, CrewAI, AutoGen | Automate leads and campaigns with Make and Flowise | Code stateful graphs in LangGraph |
| M5 Error Mitigation | Loop detection, failure recovery, cost limits | Set budgets and escalation rules | Add loop and cost guards |
| M6 Human in the Loop | Approval gateways, dual-key escalation | Design approvals and accountability | Implement approval gateways |
Course 403: Production Deployment: Cloud, Edge and Local Systems
- Summary
- Reliable enterprise deployment across cloud and local hardware.
- Prerequisites
- 402
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- LLMOps budget and vendor-agreement audit
- Labs project
- LLM service deployed with monitoring on cloud and edge
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Serving Engines | vLLM, TensorRT-LLM, continuous batching | Understand throughput and cost | Serve a model with vLLM |
| M2 Local and Edge Systems | Ollama, llama.cpp, ONNX, NPU acceleration | Decide between cloud and device | Run models locally and on NPUs |
| M3 Hybrid Routing | Cost-optimized routing, latency balancing | Manage usage budgets | Build a routing layer |
| M4 LLMOps Tracking | LangSmith, Phoenix, OpenTelemetry tracing | Track usage and quality | Add tracing and evaluation |
| M5 Privacy Architecture | Differential privacy, PII filters, local guardrails | Audit API and data agreements | Implement PII filtering |
| M6 Sustainable Compute | GPU orchestration, green compute, compression | Account for energy and cost | Compress and orchestrate workloads |
Track 5
Future Horizons, AGI and Physical Reality (2027 and beyond)
Objective: Design for generalization, embodied intelligence, and strict regulation. Level: Advanced.
Course 501: Path to AGI and Recursive Self-Improvement
- Summary
- Self-improving architectures and the theoretical paths to autonomy.
- Prerequisites
- Track 4
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- AGI-readiness strategy for your organization
- Labs project
- Synthetic data loop with verifiable tasks
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Defining AGI | Operational metrics, competence frameworks, unseen-task benchmarks | Define what readiness means for your sector | Design an evaluation suite |
| M2 Synthetic Data | LLM-in-the-loop synthesis, verifiable seed generators | Understand data limits and risks | Build a verifiable synthetic data loop |
| M3 Self-Improving Systems | Code-generation loops writing optimizers | Plan for workforce impact | Prototype a self-improvement loop |
| M4 Automated Science | Chemical discovery engines, robotic lab pipelines | Spot industry-disruption scenarios | Sketch a discovery pipeline |
| M5 Alignment Theory | Post-AGI control, scalable oversight | Ask the right governance questions | Study oversight protocols |
| M6 Long-Term Projections | AGI timeline analyses (estimates, 2027-2033) | Plan scenarios and workforce adaptation | Critique timeline forecasts |
Course 502: Embodied Intelligence and Humanoid Robotics
- Summary
- Where models meet the physical world.
- Prerequisites
- Track 4
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- Automation infrastructure plan for logistics or warehousing
- Labs project
- VLA policy trained in simulation
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Vision-Language-Action Models | Perception translated to actuator control | Identify embodied use cases | Build a VLA policy |
| M2 Kinematics and Control | Trajectory optimization, torque control, sensorimotor loops | Understand physical limits | Optimize trajectories |
| M3 Simulation to Real | Physics simulators, domain randomization | Estimate pilot costs | Train in simulation and transfer |
| M4 Robot Interaction Design | Tactile, audio, and depth sensor fusion | Design human-robot workflows | Fuse multimodal sensors |
| M5 Humanoid Control | Bipedal balance, navigation, manipulation | Evaluate humanoid readiness | Implement balance control |
| M6 Real-World Edge Cases | Variable environments, recovery routines | Plan safety and fallback | Build recovery behaviors |
Course 503: Global Policy, Alignment Safety and Compliance Engineering
- Summary
- Keeping high-risk AI compliant with tightening global rules.
- Prerequisites
- Track 4
- Duration
- 6 modules over 6 weeks, plus final assessment
- Studio project
- EU AI Act compliance plan and governance audit
- Labs project
- Audit pipeline with guardrails and explainability
| Module | Core topics | Studio (Builder) outcome | Labs (Engineer) outcome |
|---|---|---|---|
| M1 Legal Frameworks | EU AI Act risk classes, global mandates | Classify your AI systems by risk | Map systems to obligations |
| M2 Algorithmic Auditing | Bias metrics, SHAP, LIME | Commission and read audits | Run bias and explainability audits |
| M3 Copyright and Data Lineage | IP risk, lineage tracking | Set media and brand-safety rules | Implement data lineage |
| M4 Safety Infrastructure | Kill-switches, outage guardrails, defense in depth | Define incident plans | Build layered guardrails |
| M5 Alignment Standards | Constitutional AI, automated policy checks | Write enforceable AI policies | Implement policy checks |
| M6 Institutional Governance | Ethics boards, compliance reporting, audit logs | Set up governance and deepfake detection rules | Build audit logging |