Integrated Project & Handover

Core SystemsRequired project

Deliver your application, demonstrate the evidence, and revise it after review.

Integrated Project

Develop one small application that solves a clearly defined problem using a trained model, a pretrained model, or an LLM. Extend it with your two electives where practical.

Choose a manageable problem

• A prediction service with a clear target and evaluated model.
• A structured text-extraction or classification service with validated results.
• A document-search assistant when choosing the RAG elective.

Write a short problem statement, identify the intended user, and define what a successful response looks like before implementation.

Required project components

• A working AI component with documented inputs, outputs, and limitations.
• A justified baseline and a fixed evaluation procedure.
• A maintained Python project with configuration, logging, tests, and a dependency setup.
• A validated API and a small browser or command-line client.
• Persistence for a suitable part of the application, such as requests, results, or job state.
• A containerized service deployed to an agreed environment.
• An architecture diagram and instructions for setup, use, testing, and evaluation.
• A review cycle showing at least one meaningful revision in response to feedback.

Evidence from the other required modules

The project may focus on one AI approach. Keep the other required exercises available in the repository or linked from its README. For example, an LLM-based final application still includes the ML model comparison and deep learning training/debugging exercise.

Integrating electives

For each of your two electives, state the feature you implemented, the outcome you evaluated, and the evidence demonstrating it. You may use one repository for everything, with clear run instructions for each component.

Final handover

Provide the repository, deployed demonstration, evaluation results, architecture diagram, and a short explanation of tradeoffs and remaining limitations. Another person should be able to reproduce your main workflow from the documentation.

Review and handover

Plan your time around implementation, debugging, review, and revision as well as reading and watching. The stage's calendar and hour allocation should be agreed for the cohort after the practice workload is sized.

Learning sequence

• Start: verify prerequisites, create the repository, and begin the engineering workflow.
• Model foundations: complete ML evaluation and deep learning, keeping results and debugging notes.
• Applications and systems: complete LLM applications and systems, then build and deploy your API.
• Depth: complete your two electives after their prerequisites; extend the integrated project where useful.
• Handover: demonstrate the application, explain the evidence, address feedback, and document the final result.

A practice cycle for each module

• Study the assigned sections and run the examples.
• Change the example or build the module task yourself.
• Test expected behavior and inspect a meaningful failure.
• Explain your choices and request feedback.
• Revise the implementation and record what improved.

Completion expectations

Complete all five required modules, two electives, and the integrated project. Resource completion supports progress; your implementations, explanations, debugging, and revisions demonstrate the learning outcomes.

• Correctness: the application and exercises behave as documented.
• Evaluation: results use suitable baselines, splits/test cases, and metrics.
• Engineering: another person can run the code, tests, and main workflow.
• Understanding: you can explain decisions and diagnose a relevant failure.
• Iteration: review feedback leads to a documented improvement.

Using AI during the stage

Use AI tools to explore, implement, test, and review. You are responsible for the final behavior and for explaining submitted work. Record one representative AI-assisted change and how you verified it; keep generated claims about results aligned with what your tests and evaluations actually show.

Resource access and scope

Use the assigned subset of each resource. Clean Code is further reading when access is available; the required engineering work can use the linked free documentation. Keep compute and deployment small, using local models, Academy-provided endpoints, or agreed infrastructure where suitable.

Workload

Use the provisional estimates where supplied. Other workloads remain pending pilot. Include implementation, debugging, review, and revision.