GitHub portfolio for AI roles: what reviewers look at

A GitHub portfolio for AI roles is judged mostly on its pinned repositories and their READMEs. A reviewer who follows the link from your resume typically looks at what is pinned, opens the most relevant project, reads the README, and skims some code and the commit history. Two to four well-explained projects with a clear evaluation section do more for you than dozens of unexplained repositories.

What do reviewers look at on a GitHub profile?

Reviewers look at a GitHub profile in a predictable order, and most of them stop early. The people who open your profile are usually engineers on the hiring team or interviewers preparing questions, and they have limited time per candidate.

Each step is a chance to keep the reviewer or lose them. If the pinned area shows nothing relevant, the review ends at step one. If the README is empty, the review ends at step three, however good the code is.

Which repositories should you pin?

Pin two to four repositories that you built yourself and that match the role you want. GitHub lets you choose up to six pinned items; filling all six with weak projects dilutes the strong ones.

A good pinned set for an AI role contains:

• One end-to-end project. Data in, model or LLM in the middle, a usable interface or API out, with evaluation.
• One project matching your target track. A retrieval application for a GenAI role, a training and serving pipeline for an ML role, an orchestration or monitoring project for a platform role.
• Optionally, one that shows engineering range. A well-tested library, a tool you actually use, or a real contribution to an open-source project.

Give each pinned repository a clear name and a one-line description, since both show on the profile page. "rag-policy-assistant: question answering over public policy documents with retrieval and an evaluation set" tells the reviewer what to expect. "project-final-2" does not.

For what to build, the guide to AI portfolio projects that get interviews covers project ideas with enough depth to discuss in an interview.

What should the README of an AI project include?

The README of an AI project should include what the project does, how it works, how to run it, how it was evaluated, and what its limits are. The README is the page a reviewer reads, so treat it as the project's cover note.

A skeleton to start from:

# Policy Assistant

Answers questions about public policy documents using retrieval and
a language model. Built to learn how retrieval quality affects answers.

## Approach
- Ingestion: splits documents by section, stores embeddings
- Retrieval: vector search, then re-ranking
- Generation: prompt with retrieved passages and citations

## Run it
pip install -r requirements.txt
python ingest.py data/
python app.py

## Evaluation
40 hand-written questions with expected source sections.
Reports retrieval hit rate and answer correctness for each version.

## Limits
Fails on questions that span several documents. Tables are not parsed.

The figures and names in the skeleton are placeholders for your own. The evaluation and limits sections are the ones most candidates skip and the ones reviewers remember. Stating where your system fails shows that you tested it. If evaluation is new to you, the guide on how to evaluate an LLM application explains test sets and baselines.

Does commit history matter to reviewers?

Commit history matters as a sign that you built the project yourself over time, not as a count. A reviewer is not grading your activity graph. A history with many small, described commits reads as real work; a single commit titled "initial commit" containing the whole project reads as a copy or an upload.

Do not rewrite history to look busier, and do not worry about gaps. Engineers with full-time jobs have quiet months, and reviewers know that.

What should you remove from your GitHub?

Remove, archive or make private anything that weakens the first impression or should never have been public. Cleaning takes an hour and changes how the whole profile reads.

A removal checklist:

• Secrets. API keys, passwords and tokens in any file or past commit. Revoke the key first, then remove it; deleting the file alone leaves it in history.
• Employer or client code. Anything written for work, including snippets, configurations and internal documents.
• Tutorial copies. Course exercises and follow-along projects identical to thousands of others. Keep them private.
• Empty and abandoned repositories. A name with no content, or a project with only a default README.
• Unexplained forks. Forks you never changed. They crowd the list and suggest activity that is not yours.
• Large committed files. Datasets, model weights and notebook outputs that make a repository slow to open.

How should code and notebooks be organised?

Organise an AI project as code in modules with a short notebook for exploration, not as one long notebook. Production AI work at a Global Capability Centre is software engineering, and a reviewer for those roles looks for the habits of an engineer.

Signals that take little effort:

• A folder layout a newcomer can follow, such as src/, tests/, data/ with a note on where data comes from, and notebooks/.
• A requirements or environment file so the project can be installed.
• A few tests for the parts with real logic, such as chunking, parsing or feature code.
• Configuration in a file or environment variables, not hard-coded paths from your laptop.
• Notebooks that run top to bottom, with outputs cleared or kept small.

You do not need continuous integration, containers and a deployment for every project. One project with these basics done properly is enough to show how you work.

How do you connect GitHub to your resume and LinkedIn?

Connect GitHub to your resume and LinkedIn by linking the profile in the resume header and linking each project directly beside its description. A reviewer should reach the right repository in one click, not search your profile for it.

On the resume, each project entry carries its own link and one line on evaluation, as described in the AI engineer resume guide for GCC roles. On LinkedIn, add your best repository to the Featured section; the guide to a LinkedIn profile for AI engineers covers the rest of the profile.

Then check how the resume and its links read together. The free resume check on The GCC Engineer evaluates a resume for AI roles at GCCs, and the free plan includes 5 resume evaluations, shared with job fit checks.

Whatever you link, be ready to explain every line of it. Interviewers commonly pick a project from the profile and ask why you made each choice.