AI and Data Developer
The programme prepares future AI developers to turn data and language models into working applications. Through practical projects, students learn to train and evaluate models, build APIs and interfaces, and develop AI assistants that use an organisation’s own information.
Education overview
Companies need developers who can assess a practical AI use case, prepare suitable data and integrate models into usable services. The programme focuses on predictive machine learning, applications built with FastAPI and Streamlit, and language-model solutions with structured outputs and retrieval-augmented generation (RAG). Students learn to demonstrate usefulness through testing and evaluation.
Relevant roles
Junior AI developer, Python application developer and data developer working with machine learning services and LLM applications.
Learning outcomes
- Prepare data with Python and SQL, explore its suitability for modelling and prevent data leakage.
- Train regression and classification models, compare algorithms and evaluate performance on unseen data.
- Use FastAPI to serve data and model predictions through validated API endpoints.
- Build Streamlit interfaces that connect user input, backend services and model results.
- Use PydanticAI and OpenRouter to integrate language models and validate structured outputs.
- Build a RAG workflow using document preparation, embeddings, a vector database and source-based answers.
- Evaluate prediction quality and LLM responses, document limitations and apply appropriate data and access safeguards.
Core technologies
Python · Machine learning · FastAPI · LLMs · RAG
Expected level of independence
Graduates should be able to build and test a bounded AI feature within a team, explain their evaluation and hand over reproducible code with review. They should recognise when data, model quality or security concerns require specialist guidance. The expected level is junior application development rather than independent responsibility for a production AI platform.
Curriculum and structure
Python and software development foundations
Main content
Variables, types, conditions, loops, functions, collections, exceptions and basic object-oriented programming. Reusable modules, virtual environments, dependency management, debugging and tests. Jupyter supports initial experiments; Git and GitHub support collaboration.
Learning outcomes
- Write and debug readable Python programs for a defined task.
- Organise reusable code and validate inputs with clear error handling.
- Use a reproducible environment, version control and basic tests to share work.
Data preparation and SQL for AI
Main content
Load data from files, APIs and relational databases. Use SQL to retrieve and combine records and Pandas to inspect types, missing values, duplicates and inconsistent labels. Prepare traceable datasets for training and document their origin and intended use.
Learning outcomes
- Retrieve and combine relevant records using SQL and Python.
- Detect data quality issues and implement repeatable preparation steps.
- Define model inputs and target variables, checking that the data is appropriate for the task.
Exploratory analysis, regression and classification
Main content
Exploratory data analysis (EDA), distributions, relationships and target definition. Train/test splits, preprocessing and baseline models with scikit-learn. Linear regression for numerical targets; logistic regression, K-nearest neighbours (KNN) and decision trees for introductory predictive tasks.
Learning outcomes
- Use EDA to identify patterns, missing information and risks before training.
- Choose regression or classification according to the problem and target.
- Fit preprocessing on training data and keep test data separate to prevent leakage.
- Train introductory models and compare them with a simple baseline.
Ensemble models, tuning and evaluation
Main content
Random Forest and gradient boosting with LightGBM, CatBoost and XGBoost. Cross-validation, hyperparameter tuning, overfitting and class imbalance. Regression metrics such as MAE and RMSE; classification metrics such as precision, recall, F1 and confusion matrices.
Learning outcomes
- Train and compare ensemble models on the same data split and evaluation criteria.
- Choose metrics that reflect the consequences of errors in the use case.
- Tune models using training and validation data, reserving test data for final evaluation.
- Explain errors, generalisation limits and whether a more complex model improves on the baseline.
FastAPI and Streamlit application development
Main content
Serve datasets and trained model predictions through FastAPI. Request and response schemas, input validation, error responses and API tests. Build Streamlit interfaces that call the backend, accept user input and present results. Connect frontend, backend and model into a complete application.
Learning outcomes
- Create validated endpoints for retrieving data and requesting predictions.
- Load a saved model with the preprocessing needed to reproduce its predictions.
- Build a Streamlit interface that communicates with the API and handles failures clearly.
- Document dependencies and configuration so another developer can run and test the application.
Language models and structured outputs
Main content
Language-model integration with PydanticAI and OpenRouter. Instructions, conversation context, schema-based outputs and validation. Build a chatbot and compare responses using representative examples. Handle invalid outputs, service errors, latency and token costs.
Learning outcomes
- Connect a language model to a Python application using PydanticAI and OpenRouter.
- Define and validate a structured response schema before using model output in application logic.
- Build a conversational interface with explicit context handling and error behaviour.
- Evaluate output quality and record failures, response times and cost trade-offs.
Responsible AI, data protection and application safety
Main content
Appropriate use of personal and confidential data, dataset permissions, access control and secure handling of credentials. Bias, unsupported claims, prompt injection and unintended disclosure. Human review, traceability and clear communication of model limitations.
Learning outcomes
- Identify data, access and misuse risks in a proposed AI application.
- Apply safeguards for sensitive inputs, credentials and restricted documents.
- Test misleading or unsafe model behaviour and define when human review is needed.
- Document intended use, limitations and the evidence supporting the solution.
RAG and integrated AI applications
Main content
Retrieval-augmented generation (RAG): document ingestion, chunking, embeddings, vector databases and retrieval. Combine retrieved passages with an LLM to answer questions using an organisation’s information. Integrate the workflow with FastAPI and Streamlit and evaluate retrieval and answer quality separately.
Learning outcomes
- Prepare documents and index their embeddings in a vector database.
- Retrieve relevant passages and produce answers with references to the supplied sources.
- Handle questions that the available documents cannot support and preserve document access boundaries.
- Deliver an integrated RAG application with evaluation questions, tests and a documented handover.
Work-based learning (LIA)
Main content
Apply Python, model evaluation or AI application development in supervised workplace tasks with agreed learning goals.
Learning outcomes
- Contribute a scoped AI or data-development task within a team.
- Use the team’s workflows, explain technical decisions and act on professional feedback.
Entry requirements and target audience
Beginners and career changers who want to build software with machine learning and language models. The focus is programming, model evaluation and application development, with data preparation taught as the foundation for those tasks.
Entry requirements
- English sufficient to follow teaching, read technical documentation and explain your work.
- High school mathematics, including algebra, percentages and interpreting graphs. Statistics and model evaluation are introduced during the programme.
- No previous programming or AI development experience is required. Python is taught from the fundamentals.
Expected starting knowledge
Students need basic computer skills and should be ready to practise programming, reason about numerical results and work with feedback. Prior knowledge of machine learning libraries, cloud services or language models is not assumed.
How students apply their knowledge
Teaching and study format
Instructor-led online teaching, guided coding exercises and individual and team assignments. Students progress from Python and data preparation to predictive models, web applications and document-based AI assistants. Git, GitHub and reproducible environments are used throughout; notebooks support exploration before code is organised into applications.
How learning is demonstrated
- Submit a reproducible modelling workflow with documented data checks, train/test separation and baseline comparisons.
- Demonstrate a FastAPI service and Streamlit interface, including invalid input and service failure handling.
- Show validated structured LLM responses and test behaviour when the model produces an unusable answer.
- Demonstrate document retrieval and source-grounded answers using a defined RAG evaluation set.
- Explain privacy, access controls, costs, known limitations and the changes made after feedback.
Would you like to endorse this education?
Do these skills match what your company needs? Your perspective helps us make the education relevant to real work.
You are not required to
- Teach or prepare course materials.
- Hire students.
- Host interns.
- Attend recurring meetings.
How you can support us
- Share your current and anticipated skills needs.
- Give brief feedback on the curriculum and learning outcomes.
- Endorse the education’s relevance based on your own assessment.
Providing your initial feedback takes no more than 1 hour.
Give feedback and endorse this education