A generative AI internship is easiest to understand by what you build. This post lists six small projects that show real skill, what each one needs, what is free, and how to evaluate the results honestly. Pick two or three, finish them and write them up. That will teach you more than a long list of tools.
Plain terms about ours. The EDWartens generative AI and AI agents internships are training, not employment, written with the Wartens AI company OFORO. The first 4 weeks are free, with no fee to join. Continuing is paid. Finishing the first 4 weeks earns a free certificate of completion (QR-verifiable), while the graded Internship Certificate and other documents are optional paid extras shown in your own currency, and there is no stipend. We promise no job, interview or placement.
Finish the first 4 weeks and you earn a free certificate of completion, verifiable by QR code, which makes this a free online internship with certificate. The graded Internship Certificate, for a whole internship, is a separate paid document.
What does a generative AI internship involve?
Most of the work is not inventing models. It is using existing language models well. That means writing clear prompts, giving a model the right information, checking what comes back, wiring a model into a small workflow and measuring whether it works. The skill employers look for is judgement: knowing when to trust an output and when not to.
Our generative AI internship starts with what generative AI is and is not, how large language models work, prompting fundamentals and advanced prompting in the first four free weeks. The AI agents internship starts with what an agent is, automation basics with n8n, connecting a model to a workflow, and tools and function calling.
Project 1: document Q&A
The idea. Ask questions about a set of documents you own, such as a manual, a policy or your notes, and get answers that quote the source.
What it shows. That you can give a model the right context and make it cite where an answer came from.
What it needs. A few documents, a chat model, and either a simple script or a no-code tool. Test it with twenty real questions and record where it makes things up.
Project 2: a summariser
The idea. Turn long text into a short summary in a fixed format, such as three bullets and one action.
What it shows. Prompt design, output format control and a sense of what to leave out.
What it needs. Ten to twenty sample texts, a prompt with a role, task and format, and a way to compare outputs. Rewrite the prompt one change at a time and keep a log of what each change did.
Project 3: a classifier
The idea. Sort messages or tickets into labels: urgent or not, type of request, topic.
What it shows. That you can use a model for structured output and measure it against labels you wrote yourself.
What it needs. About fifty labelled examples. Report how often the model agrees with you, show the mistakes, and note which labels confuse it.
Project 4: a simple agent
The idea. A small agent with one or two tools, for example a calculator and a lookup, that decides which to use to answer a question.
What it shows. That you understand the loop of plan, act, observe and the need for guardrails.
What it needs. A model that supports tool use, clear tool descriptions and a hard limit on steps. Test whether it picks the right tool, and what happens when a tool fails. Keep it small: a safe agent beats a clever one.
Project 5: an evaluation sheet
The idea. A spreadsheet that scores model outputs against a few simple criteria: correct, complete, clear, safe.
What it shows. The habit that separates a demo from a tool: measuring. It also shows you can be honest about quality.
What it needs. A set of prompts, outputs from two prompt versions or two models, and a short scoring guide. Even a ten-row sheet makes you look careful.
Project 6: a small app
The idea. Wrap one of the above in a simple page or form so someone else can use it.
What it shows. That you can finish and share something, not only run it in a notebook.
What it needs. A basic web form or a no-code builder, plus a short guide for users. Keep keys out of the code and out of any public repository.
Free tools and a cost cap
Many models offer limited free use, and open models can run on your own computer. Free offers change, so check current terms before you rely on one. Hugging Face offers a free LLM course that expects good Python (page read 17 October 2026), and our free local LLMs with Ollama course shows how to run a model privately.
Set a cost cap before you start: a maximum you will spend, or a rule to use only free allowances. Check usage after each session, and never put API keys in public code. A surprise bill is a common beginner mistake.
How to evaluate outputs honestly
A model that sounds confident can still be wrong. Three habits keep you honest.
- Use a fixed test set. Write the questions before you build, so you cannot tune to them.
- Score by hand first. Read every output for a small set before you trust any automatic score.
- Record failures. Keep a list of outputs that were wrong and why. That list is the most useful page in your write-up.
Showing prompts, results and failures
In your write-up, show the prompt you used, three good outputs, three bad outputs and what you changed. Include the date and the model name, because results change between versions. Hiding failures makes work look weaker, not stronger. Our guide to putting an online internship on your resume shows how to link this kind of proof.
Free training internship versus paid roles
Employer roles in generative AI exist but are competitive and often want prior projects. A training internship gives you structure and a place to build the first ones. Ours is free for the first 4 weeks, and the free internship page spells out what is paid afterwards. For more on how the routes compare, read the AI internship for students guide and our post on getting an AI internship with no experience.
Which projects to pick first
You do not need all six. A sensible order for a beginner is the summariser first, because it teaches prompt design with little setup, then the evaluation sheet, which teaches you to measure. Add document Q&A or the classifier next, depending on whether you prefer text or labels. Leave the agent and the app until you have two finished projects.
If you are not an engineer, pick projects from your own field. An accountant might summarise invoices, an HR officer might classify candidate queries and a project manager might draft status reports. The skill is the same: clear instructions, careful testing and honest results.
What a good write-up includes
- The goal in one sentence and the model used, with the date.
- The prompt, shown in full.
- Three good outputs and three bad ones, with a note on why.
- The score from your test set and how you measured it.
- What you would change next.
Keep it to a page or two. A reader should see that you tried, measured and learned, and should find it easy to repeat what you did.
Safety and data basics
- Never paste private, customer or health data into a public tool.
- Remove names and numbers from sample documents.
- Tell readers that outputs are drafts that a person must check.
- Do not present a model's answer as advice about medicine, law or money.
- Keep keys and passwords out of code you share.
A note on honesty about limits
Language models can sound sure and still be wrong. State this plainly in every project you publish. Say which model and version you used, on which date, and with what settings. State that results can change when the model changes. A short limits paragraph makes your work more trusted, not less.
It also protects you. If someone reruns your project and gets a different answer, your write-up already explains why.
Next steps
Pick two projects, one no-code and one with Python, and finish them within a month. Write them up with their failures, and link them. If you want weekly structure, look at the AI internships list, then choose a start date on the planner. The AI and machine learning internship is a good companion if you want the model basics behind these projects.
Practise with us
Internship
Generative AI and Prompt Engineering
First 4 weeks free. Weekly briefs, a work record and documents you can verify if you choose.
See the internshipRoles
Generative AI
Roles that ask for this skill, 60 open in this country now. Free to see and to apply. We are not a recruitment agency.
See rolesTake the free course
FreeAI and machine learning · Beginner · Free
Generative AI and LLM Foundations
FreeAI and machine learning · Beginner · Free
Prompt Engineering for Engineers
FreeAI and machine learning · Beginner · Free
AI Agents and Workflow Automation with n8n
FreeAI and machine learning · Intermediate · Free
Private AI with Local LLMs: Ollama, Open WebUI and Offline RAG
Questions
Do I need to code?
For some projects, no. You can build a prompt library, an evaluation sheet and a no-code workflow without code. For document Q&A, a classifier or a small app, basic Python helps. Start with the no-code ones and add Python as you go; the order is up to you.
Which model is free?
Free offers change often, so check each provider's current terms before you plan around one. Many give limited free use, and open models can run on your own machine. Set a cost cap before you start, and use the free allowance for testing so a surprise bill never arrives.
Is prompt engineering a job?
It is a skill more than a job title. Writing clear prompts, testing them and judging the outputs helps in many roles, from analyst to engineer to marketer. Treat it as one part of a wider skill set, alongside checking facts and understanding the data you give a model.
What is an AI agent?
An agent is a program in which a language model decides which step or tool to use next, in a loop, toward a goal. A plain automation follows a fixed path; an agent chooses. Start with a small one that has one or two tools, and test it carefully.
Can non-engineers join?
Yes. The skills of writing clear instructions, testing and checking results matter in finance, HR, marketing and project management too. We offer separate AI internships for those fields, so you do not need to code to start.
How long should I spend?
Plan around one weekend for each small project, or about a week at a steady pace. Six projects fit in a couple of months alongside study. A finished small project with an honest write-up is worth more than a large unfinished one.
Sources
- Hugging Face, LLM Course (read 17 October 2026)
- Google for Developers, Machine Learning Crash Course (read 17 October 2026)
- EDWartens, Generative AI internship: the weekly course (read 18 October 2026)
Written by the EDWartens engineering team for general education. Product names are trademarks of their owners; mentioning them does not imply endorsement. Prices and terms of other providers were checked on the date shown and can change.
Free Generative AI and Prompt Engineering Courses With Certificate, Compared
Generative AI for Engineers: What LLMs Can and Cannot Do for Technical Work
Work From Home Internship for Students: Real Options
Data Science Internship for Beginners: A 4-Week Plan
SCADA Internship Online: What You Build in 4 Weeks
1Z0-1122 Study Guide: OCI AI Foundations, Topic by Topic