Defined condition
If the name contains “invoice”, move the file to the invoice folder.
Progressive path
From words to working systems.
This intermediate level builds a bridge between intuition and implementation: a map for reading data, models, algorithms and automation without losing the thread.
The same problem, different tools
Organising documents: rule, RPA, machine learning or generative AI?
The tools can work together but answer different needs.
Operational sequence
Open the mailbox, save the attachment, rename it and update the register.
Learned classification
Classify the document from text and structure even when its name is unclear.
Draft for review
Prepare a content summary and an initial response subject to human review.
AI, ML and algorithms
Three connected circles, not three synonyms.
Artificial intelligence is the broad field: it includes systems that classify, estimate, search or generate. Machine learning is the part of AI where a rule is learned from data. An algorithm is a defined sequence of steps and may be statistical, deterministic or learned.
The question comes before the tool: describing, forecasting, classifying, grouping, detecting anomalies or setting priorities call for different method families.
Data types
The shape of data guides the method.
Tables
Rows, columns, categories and measures: prices, sensors, KPIs, compositions and accounts.
Time series
Values ordered in time. Rhythm, trend, seasonality, regime and change become central.
Text and documents
Reports, notices, articles and PDFs require parsing, retrieval, synthesis and source checks.
Images and maps
Pixels, satellite bands and coordinates describe shape, matter, temperature and space.
Graphs and networks
Nodes and relations represent value chains, dependencies, citations, flows or connections.
Events and logs
A sequence of discrete facts tells the story of triggers, exceptions, transitions and system state.
Methods and questions
Each method answers a specific question.
The technical name follows the concrete question.
| Family | Plain question | Everyday or applied example |
|---|---|---|
| Rules | Is the condition true? | Flag an expired document. |
| Regression | What might the value be? | Estimate consumption. |
| Classification | Which category does it belong to? | Route a request. |
| Clustering | Which cases are similar? | Group behaviour profiles. |
| Anomaly detection | What is unusual? | Identify an abnormal sensor. |
| Time series | How does it change over time? | Estimate demand or temperature. |
| Neural networks | Which complex structure is present? | Recognise elements in images or text. |
Machine-learning cycle
Training is not the same as demonstrating.
- Define the target: what should be estimated and when it becomes observable.
- Build features: representations available at decision time.
- Separate time: training, validation and test without leakage.
- Compare a baseline: the model must add value over a simple rule.
- Measure uncertainty: errors, calibration and stability by period.
- Monitor: data, drift, failure modes and matured outcomes.
Data automation and RPA
Value emerges when the process remains readable.
Trigger
Schedule, event, new file, updated API or manual command.
Acquire
Retries, timeouts, source date, licences and provenance.
Control
Schema, units, missing values, ranges and freshness.
Compute
Rules, statistics, ML, geospatial analysis or document composition.
Decide
Publish, skip, block, degrade or request review.
Trace
State, artifacts, logs, hashes and URIs make the route reconstructable.