The AI Portfolio: What to Build When You Haven’t Had Your First Job

An AI portfolio can help an employer answer a question that a CV cannot: what can this person actually build?

That matters when you have not yet had your first professional role. Without a job title or employer reference, it can be difficult to demonstrate how you would contribute. A thoughtful portfolio creates another form of evidence. It shows how you understand a problem, design a workflow, test the result and communicate what you learned.

The goal is not to fill a website with every experiment you have tried. It is to present a small number of complete projects that resemble useful business work. Each project should have a clear user, realistic constraints and a visible outcome.

You can build with simulated businesses and invented data. Be transparent about that. Employers are not expecting a personal project to prove that you have managed a live enterprise system. They are looking for signs of judgement, initiative, attention to detail and the ability to finish.

A good portfolio makes your level clear while showing your potential. It gives an interviewer something concrete to discuss and gives you a stronger story than simply saying that you are interested in AI.

What your portfolio needs to prove

Your portfolio should prove more than familiarity with tools. It should show that you can connect technology to a useful outcome.

An employer should be able to identify the problem, user, workflow and result. They should see where AI adds value, where ordinary rules are more appropriate and where a person remains responsible.

The portfolio should also reveal how you dealt with uncertainty. AI outputs are not automatically correct. Show how you tested them, which failures you discovered and how the design changed.

Finally, prove that you can communicate. Use concise summaries, diagrams and demonstrations. If a reviewer cannot understand the project without opening the automation platform, the case study is doing too little work.

Think of each project as evidence for several abilities: problem framing, building, testing, responsibility and explanation.

Project one: an enquiry triage assistant

Create a fictional service business that receives enquiries through a form or shared inbox. The system should identify the enquiry type, extract important details and route it to the correct team.

Use several categories, such as sales, support, billing and unsuitable requests. Add rules for missing information and uncertain classifications. A person should review anything sensitive or ambiguous.

The AI component can interpret free text and produce structured fields. The automation component can add the result to a table, prepare an internal summary and notify the correct owner.

Test the system with clear, messy and misleading enquiries. Record accuracy and escalation behaviour. This project demonstrates text interpretation, structured output, workflow design and human handover.

In the case study, explain that it is a prototype using invented data. Show how the same pattern could apply to professional services, education, property or healthcare administration, while recognising that regulated contexts need additional controls.

Project two: a sales research workflow

Build a workflow that turns an approved company name or website into a concise research brief for a salesperson.

Define which public sources are acceptable and which facts matter. The output might include the organisation’s offer, likely customers, relevant recent signals and questions for a discovery call. Require source links and prevent the model from filling gaps with assumptions.

Add a review step before anything enters a customer system. The project should support preparation, not automate unsolicited contact or invent personal information.

Test companies with clear websites, limited information and similar names. Check whether the system distinguishes fact from inference. Measure whether the brief follows a consistent structure.

This project demonstrates research, source handling, commercial context and responsible boundaries. It is stronger than a generic “lead generation agent” because it shows how a user would trust the result.

Project three: a grounded knowledge assistant

Create a small knowledge base for a simulated organisation. It might contain service descriptions, policies, opening hours and escalation information. Build an assistant that answers questions only from those approved materials.

The assistant should cite or identify its source, admit when the answer is unavailable and direct appropriate questions to a person. Include instructions for conflicting or outdated information.

Test ordinary questions, ambiguous wording, requests outside scope and attempts to make the assistant ignore its rules. Record how often it gives a grounded answer or an appropriate refusal.

This project demonstrates context design, information retrieval, testing and safety. It also gives you an opportunity to discuss document quality. A knowledge assistant cannot be more reliable than the information supporting it.

In your portfolio, show an example that succeeded and one that required escalation. That comparison makes the design easier to understand.

Project four: an operational reporting workflow

Use a sample dataset to produce a weekly report for a fictional operations manager. The workflow might calculate fixed metrics, identify changes and ask an AI model to draft a plain-English commentary.

Keep calculations deterministic. The model should explain verified figures, not calculate them from memory. Pass the numbers into a clear template and ask the model to identify patterns without inventing causes.

Add checks for missing rows, unexpected values and reporting periods. Require a person to approve the narrative before distribution.

This project demonstrates the important difference between computation and interpretation. It also shows that you can design for recurring use rather than a one-off response.

Include the input data, calculation logic, report example and test notes. Explain how the workflow might save preparation time while leaving accountability with the manager.

Project five: a content repurposing system

Create a system that turns one approved source, such as a webinar transcript or article, into several draft formats. The purpose is to preserve the source while reducing repetitive rewriting.

Define the audience, tone and rules for each format. Require every claim to remain grounded in the original material. Add a checklist for unsupported statements, repetition and brand suitability.

The workflow might prepare a newsletter summary, social post and short sales-enablement note. All outputs should remain drafts for human review.

This project demonstrates structured content operations rather than uncontrolled generation. It shows that you understand approval and source fidelity.

Avoid presenting the system as a machine that replaces creative judgement. Explain how it gives a person a stronger starting point and allows them to spend more time refining the message.

How to document every project

Use a consistent structure so employers can compare your work. Start with a short summary, then describe the user, current process and desired outcome.

Add a workflow diagram. Label triggers, data, rules, AI steps, approvals and outputs. Include screenshots only when they help explain a decision.

Describe the test set and results. Mention failures and changes. State the limitations and what would be required before live use, including data protection, access, legal or security review where relevant.

Provide a brief video demonstration with captions or clear narration. Keep sample information visible and remove secrets or credentials.

End with a reflection. Explain what you would do differently and what skill the project helped you develop. Honest reflection makes a junior portfolio more credible.

How to publish and share it

A simple portfolio site is enough. Give every project its own page and make the summary understandable on a phone. You can also provide downloadable case studies or a public document.

Use clear titles based on outcomes, such as “Customer enquiry triage assistant”, rather than internal names. Include the date and note which tools were used, but do not let the tool list dominate.

Link to the most relevant project from each application. A customer operations role should receive the knowledge or triage example. A sales operations role should receive the research workflow.

Protect your own work. Remove access keys, private data and editable endpoints. If a demonstration has running costs, use a recorded video rather than leaving it publicly open.

The portfolio exists to begin a useful conversation. Make it easy for someone to understand your contribution and ask you better questions.

FAQs

Do I need to publish the actual workflow?

No. A demonstration, diagram and detailed case study may be enough. Never expose credentials, private data or an endpoint that could be abused.

Can all my projects use simulated businesses?

Yes, especially before your first role. Make the scenarios realistic and state clearly that the data and organisation are fictional.

Should I include the tools I used?

Yes, but keep them secondary. Employers care more about the problem, decisions, tests and outcome than the logo collection.

How polished should the portfolio be?

It should be clear, accurate and easy to navigate. Do not delay publication for elaborate design. Strong evidence and explanation matter more than visual effects.

Build a portfolio employers can inspect

Working Fellowship participants build a portfolio of applied AI projects and a supervised capstone, with direct feedback and formal assessment. The result is evidence of practical ability rather than a certificate for watching lessons.

Explore the Certified AI Implementation Practitioner programme and apply at workingfellowship.com.