AI engineering · mapped from public work
Map what
you’ve built.
A shareable, evidence-backed profile of your AI work, for recruiters, collaborators and networking, or just to stay competitive in the AI space.
AI-Atlas reads your public GitHub projects and maps the AI capabilities your work actually demonstrates — judged on what you built, not on what you imported. Ranked only when claimed.
- Public repositories only
- No account needed to map
- Code is never executed
- Your dataWhat happens to my data?
Live Atlas coverage
Pop Alexandru
@pop123-ux · 🇷🇴
AI-Atlas score
71.1/100
Specialist
- #4= global
- #1 in Romania
- ML73
- Vision69
- NLP & LLMs67
- Data42
- Agents33
- MLOps14
Projects behind it
- ORBITLibrary / Framework67.4
- ORBIT-allTraining Experiment66.0
- AlexNet-from-scratchArchitecture Reimplementation60.5
Mapped
Open this AtlasHow it works, on a real Atlas
0101Paste a GitHub username
AI-Atlas finds the public repositories, sets aside forks and empty ones, and reads the code of those that do AI work. Nothing is installed or run.
0202An AI judge scores each project
Each project is scored against a weighted rubric, and the claims in its README are checked against the files the judge read. The owner sees every line of it.
- 0302
Plan your next move
Your Atlas, scores and leaderboards are free. Planning credits turn a project’s evidence into a prioritized roadmap, predict what improving one rubric dimension would do to its score, or rescan a single project you pushed to. Your first roadmap is free.
0403Share your Atlas and your work
One link with your score, your strongest field, the projects behind them and, once you claim it, your place on the global and country leaderboards.
Six domains of evidence
Each domain is scored on its own, from evidence in the code you published.
01. Machine Learning
Model architectures, training loops, optimisation, classical and deep learning method work — the mathematics and mechanics of learning systems.
02. Computer Vision
Image and video understanding: classification, detection, segmentation, generative vision, multimodal perception and vision architectures.
03. Data Science
Data acquisition, cleaning, feature engineering, statistics, analysis, experimentation and visualisation that supports decisions or models.
04. NLP & LLMs
Language modelling, transformer architectures, tokenisation, fine-tuning, retrieval augmented generation, evaluation and inference of LLMs.
05. AI Agents
Tool-using and autonomous systems: planning, orchestration, memory, multi-agent coordination, function calling and agent evaluation.
06. MLOps
Getting models into and keeping them in production: serving, pipelines, experiment tracking, reproducibility, monitoring and infrastructure.
Evidence, not buzzwords.
One real project, exactly as AI-Atlas assessed it: what was read, what the code confirmed, what the judge marked down, and the score it earned, all from one commit.
- 01 · Read
The files that explain the project
- src/model.py
- tests/test_model.py
- pyproject.toml
- README.md
- src/__init__.py
Read as text. Nothing is installed or run.
- 02 · Check
Code confirms the work is real
Before any model gives an opinion, AI-Atlas checks the code itself for the work the project claims.
A model trained in PyTorch, found in
src/model.pyClaims the code does not back up don't count, whatever the judge says.
- 03 · Judge
Judged on substance, flaws included
- Technical depth55
- Conceptual difficulty50
- Originality30
- Completeness85
- Usefulness55
- Documentation70
- Reproducibility80
“The custom ReLU implementation is a trivial torch.clamp wrapper, contributing little to technical depth.”
“The README claims training on Imagenette for 40 epochs … but the repository does not show executed notebook cells or training logs to verify these results.”
- 04 · Score
One merit score, traced to its evidence
60.5merit · Advanced
AI-Atlas, not the judge, turns the assessment into a score. It counts most towards:
- Computer Visionrelevance 0.9
- Machine Learningrelevance 0.8
Designed for trust.
Your code is read, never run.
AI-Atlas documents its methodology, shows public project summaries, and gives verified owners the detailed evidence behind every score.
read: public reposNo private code requestedmode: read as textNothing installed or runrubric + cited filesOpen to verified ownersleaderboard: opt-inScanning never ranks anyoneYour data, plainly
What we read
- Your public profile and repos. Nothing private.
- No account needed to map.
- Signing in is read-only: profile and email.
What we keep
- File paths, a code outline and the scores.
- No copy of your source code.
- Your email, used only if you say yes.
What's public
- Your score, summaries and strengths.
- Limitations only if you switch them on.
- The reasoning behind scores stays with you.
How to remove it
- Revoke access in GitHub Settings → Applications.
- Remove my Atlas on your profile, or ask us.
- Never sold.
Every detail, including payments and cookies, is in the Privacy Policy.
