07 — Applications
Guiding question: “Why are we solving this problem, and what does success look like in the real world?”
This layer captures real-world use cases, business functions, and domain verticals — the problems that ML and AI systems are built to solve. Notes frame the problem, describe the users and stakeholders, outline the data reality, and specify what success means in business terms.
This layer does not contain cross-layer executable reference architectures (→ 08_implementations), concept-specific model implementations (→ home layer alongside the concept), model theory (→ 03_modeling), or engineering patterns (→ 05_ml_engineering, 06_ai_engineering).
Sublayers
01 — Prediction and Forecasting
Predicting future quantities: demand, churn, risk scores, customer lifetime value.
02 — Recommendation and Ranking
Personalised content and product discovery, feed ranking, search result ordering.
03 — Detection and Monitoring
Fraud detection, anomaly detection, quality monitoring, alerting systems.
04 — Classification and Decisioning
Binary and multi-class decisions with regulatory and operational constraints: credit, triage, routing.
05 — Generation and Assistance
LLM-powered generation: document summarization, code generation, writing assistance.
06 — Search and Retrieval
Semantic search, enterprise knowledge retrieval, RAG-based assistants.
07 — Multimodal Systems
Computer vision, document intelligence, audio-visual systems.
08 — Domain Verticals
Industry-specific applications: insurance, finance, health, e-commerce, mobility, operations.
Cross-layer Role
03_modeling/ → what algorithms apply
05_ml_engineering/ → how the system is built and operated
06_ai_engineering/ → if LLMs or foundation models are involved
07_applications/ → WHY we build it, WHO uses it, WHAT success looks like
08_implementations/ → HOW cross-layer executable patterns and end-to-end examples are implemented
09_projects/ → actual execution instances
Application notes are the business specification layer. They inform engineering decisions in all implementation layers.