The "newly open sourced" AI bootcamp has been MIT-licensed and public since March 18, 2026, and it is 523 lessons across 20 phases

A 33-second vertical clip at 1080x1920, posted 2026-10-03 by @buildwithneej (channel name "Neej"), running on original sound by Neej, sitting at 3,565 views, 206 likes, 11 comments, 31 reposts and 235 saves at capture. I analyzed 17 extracted frames and a 130-word transcript. The frames alternate between a talking-head shot in front of a bookshelf and full-screen blue-on-white motion graphics styled as pages from a technical reference manual, each labeled FIG_000 through FIG_007. Frame 1 is a screen recording of a dark-mode GitHub README showing the tabs README / Code of conduct / Contributing / MIT license, a banner reading "AI ENGINEERING FROM SCRATCH." with the subhead "20 PHASES · 523 LESSONS · 396 SKILLS · 99 PROMPTS · CERTIFICATION PREP · CLAUDE · MCPA", a twelve-language translation row, shields.io badges reading license MIT, stars 63k and web aiengineeringfromscratch.com, a SerpApi sponsor banner, a blockquote reading "84% of students already use AI tools. Only 18% feel prepared to use them professionally" followed by "523 lessons. 20 phases. ~342 hours. Python, TypeScript, Rust, Julia," and a stats line reading "114,584 readers · 181,995 page views in the last 30 days · as of 2026-08-29". Frame 3 shows an animated counter resting on 60,769 STARS. Frame 4 shows cards reading 523 LESSONS and 19 PHASES. Frame 8 is a mock Claude Code terminal running npx skills add rohitg00/ai- before the text clips. Frames 9 and 10 show a 10-QUESTION QUIZ card filling with checkmarks. Frame 13 shows four diagram cards labeled PROMPT, SKILL, AGENT and MCP under the title "Walk away with what you built." Frame 17 is the call to action: comment COURSE and the creator sends the repo.

The repo is rohitg00/ai-engineering-from-scratch, and the video never says so out loud

The voiceover calls the project "AI engineering from scratch" and stops there. The only identifying string anywhere in the video is the partially typed command in frame 8, npx skills add rohitg00/ai-, which clips before the repo slug finishes. The GitHub repository resolves that: the full slug is rohitg00/ai-engineering-from-scratch, and the README install instruction gives the complete command as npx skills add rohitg00/ai-engineering-from-scratch. The owner account rohitg00 belongs to Rohit Ghumare, a UK-based engineer whose GitHub profile lists Google Developer Expert, CNCF and Docker Captain credentials across 320 public repositories and 6,702 followers. The LICENSE file names him directly: "Copyright (c) 2026 Rohit Ghumare."

Attributing the work matters more here than usual, because the video's framing ("someone open sourced") and its gated call to action ("comment COURSE and I'll send you the repo") both route attention through the creator rather than the author. The repo is public. Nobody needs to comment to get it.

"Over 60,000 stars" undercounts, and the video contradicts itself three times in 33 seconds

The voiceover says "ripping over 60,000 stars now." The animated graphic in frame 3 settles on 60,769. The shields.io badge in the screen recording in frame 1 reads 63k. The TikTok caption says "over 62,000 stars on GitHub." Those are three different numbers inside a single post.

The GitHub API returned 63,064 stars for the repository when I checked on 2026-10-03, the day the video went up, alongside 10,787 forks and 395 watchers. The README badge in the creator's own screen recording was the accurate one. The hand-built counter graphic was roughly 2,300 stars stale, and the voiceover rounded down from there. The error direction is in the project's favor, so it reads as sloppy rather than inflated, but a number recited on camera should match the number on the screen behind it.

The phase count on screen is wrong and the lesson count is exactly right

Frame 4 shows two cards: 523 LESSONS and 19 PHASES. The voiceover says "over 500 lessons across 20 phases." The on-screen graphic is the one that got it wrong.

The repository has 20 directories under phases/, numbered 00-setup-and-tooling through 19-capstone-projects. The README's own shields badge reads phases-20. The likely cause of the graphic's error is reading the highest directory number, 19, instead of counting from zero. The voiceover's 20 matches the repo.

The lesson figure holds up under direct inspection. I pulled the full recursive git tree for the main branch (9,775 entries, untruncated) and counted directories matching the pattern phases/NN-name/NN-lesson/. The count is exactly 523, distributed unevenly: 12 lessons in Phase 0, 54 in Phase 14 (Agent Engineering), and 85 in Phase 19 (Capstone Projects). "Over 500 lessons" is accurate and conservative.

Four languages, and three of them are thin

The voiceover says "in Python, TypeScript, and Rust." The caption repeats the same three. The README blockquote visible on screen in frame 1 lists four: "Python, TypeScript, Rust, Julia." The project's About page also lists four. The video dropped Julia.

The more useful correction is the distribution. Counting source files under phases/ in the git tree returns 660 .py files, 129 .ts files, 20 .jl files and 10 .rs files. Across 523 lessons, Rust appears in roughly ten code files. Anyone picking this curriculum expecting a Rust track comparable to the Python track will find Python carrying the curriculum and Rust appearing as occasional illustration. The repo's own code/ directory structure confirms the pattern, with TypeScript the only second language present at scale.

The tutor loop is real, and it works the way the video describes

This is the part the video gets right in detail. Line 141 of the README states: "A ten-question placement quiz maps what you already know to a starting phase and saves a personalized study plan to LEARNING.md. From there, the learn skill teaches one lesson per session: concept, math, code, quiz."

The repository ships ten agent skills under skills/: start-learning, find-your-level, learn, course-guide, check-understanding, learn-mcp, learn-agent-skills, build-project, claude-certification and mcpa-certification. The README documents host-specific invocation for Codex (start-learning) and Claude Code (/start-learning). find-your-level is described as the "ten-question placement quiz" that "produces a personalized path with hour estimates," and check-understanding <phase> is an eight-question per-phase quiz. Every claim in the video's middle section (10-question quiz, personal study plan, one lesson at a time inside your terminal) maps to a documented skill.

The "walk away with what you built" claim is also supported. The README states that every lesson ships a reusable artifact: a prompt, a skill, an agent or an MCP server. Frame 13's four diagram cards (PROMPT, SKILL, AGENT, MCP) mirror that sentence directly.

It has been open source from the first commit, not open sourced recently

"Someone open sourced an entire AI engineering bootcamp" implies a recent release. The commit history says otherwise.

The repository was created on 2026-03-18 at 18:38 UTC. The first commit on main landed at 15:19:10 UTC the same day, and its message reads "Initial commit — Add MIT license and .gitignore for Python, TypeScript, Rust, Julia, notebooks." The MIT license was in the repository before any lesson content was. The second commit, two minutes later, added a README describing "19 phases, 200+ lessons." The project went public with the license already attached and has accumulated 1,813 commits on main since, with the most recent landing 2026-10-02, one day before the video.

Third-party coverage records the growth curve along the way: a dev.to post by the author describes 435 lessons, and earlier writeups cite 428 and 490+. The curriculum went from 200+ lessons to 523 over roughly six and a half months. No single open sourcing event happened. The repo grew in public.

The license status itself survives scrutiny, which is worth stating because "free course" repos frequently do not. The LICENSE file contains the standard MIT text. The GitHub API returns spdx_id: MIT. The About page states the project is "MIT-licensed and free forever," with "no token, no course upsell, and no gated content." This is an OSI license on the actual content, not a free-to-read landing page.

"Paid bootcamps charge thousands" checks out, with a real number attached

The video asserts that paid bootcamps charge thousands of dollars without naming one. Checking a first-party source rather than an aggregator: Virginia Tech's AI and Machine Learning Bootcamp lists standard tuition of $9,995 for a 23-week part-time program, discounted to $4,995 when paid upfront. That puts the comparison in the right order of magnitude for a single accredited-university-branded program, against a repository estimating ~342 hours of material at $0.

The comparison is not apples to apples. A bootcamp sells cohort scheduling, instructor access, career services and a credential. This repo sells none of those, and its certification tracks are prep material for Claude and MCPA exams rather than certificates it issues itself.

Key Takeaways

The repo is github.com/rohitg00/ai-engineering-from-scratch by Rohit Ghumare, named nowhere in the video except a half-typed terminal command in frame 8.
Correction, star count: the voiceover says "over 60,000," the graphic shows 60,769, the caption says "over 62,000," and the README badge in the same video shows 63k. The GitHub API returned 63,064 on 2026-10-03.
Correction, phase count: the on-screen card reads 19 PHASES. The repo has 20 phase directories, numbered 00 through 19, and the README badge reads phases-20. The voiceover's "20 phases" is the correct one.
Verified independently: 523 lesson directories counted from the untruncated recursive git tree of main (9,775 entries), matching the README badge exactly.
Correction, languages: the video says three languages. The README and About page say four, adding Julia. File counts under phases/ are 660 Python, 129 TypeScript, 20 Julia, 10 Rust, so Rust is incidental rather than a parallel track.
Correction, chronology: the repo has been MIT-licensed since its first commit on 2026-03-18 at 15:19:10 UTC, 199 days before the video. It launched at "19 phases, 200+ lessons" and grew to 523 lessons over 1,813 commits.
The license is genuine MIT per the LICENSE file and the GitHub API spdx_id, and the About page states "no token, no course upsell, and no gated content."
The tutor claims are accurate: a ten-question placement quiz via the find-your-level skill, a plan written to LEARNING.md, and one lesson per session via learn, installed with npx skills add rohitg00/ai-engineering-from-scratch.
Repo scale as of 2026-10-03: 63,064 stars, 10,787 forks, 395 watchers, 67 open issues, 22 contributors listed by the API, 12 translated landing pages.
The call to action gates a public MIT repository behind a comment. The URL above requires no comment.
Unverified: the "~342 hours" estimate carries no published methodology; the "84% of students already use AI tools, only 18% feel prepared" statistic visible on screen is uncited in the README; the "114,584 readers · 181,995 page views" figures are self-generated from the project's own site/stats.json as of 2026-08-29; and I could not determine what date the 60,769 star graphic was captured.

Resources

rohitg00/ai-engineering-from-scratch on GitHub — confirms the repo slug, the MIT license, 63,064 stars, 10,787 forks, creation date 2026-03-18, and the 20 directories under phases/.
The repository README — confirms the 523-lesson and 20-phase badges, the four-language list including Julia, the ~342 hour estimate, the ten-question placement quiz writing to LEARNING.md, the ten shipped agent skills, and the uncited 84%/18% statistic.
The repository LICENSE file — confirms standard MIT text, "Copyright (c) 2026 Rohit Ghumare."
Commit history for main — confirms the initial commit of 2026-03-18T15:19:10Z adding the MIT license, the same-day README describing "19 phases, 200+ lessons," and 1,813 total commits.
AI Engineering from Scratch, About page — confirms authorship, 20 phases, the four languages, and "MIT-licensed and free forever... no token, no course upsell, and no gated content."
Build It, Then Use It, by Rohit Ghumare on dev.to — confirms an earlier 435-lesson state of the curriculum and the npx skills add install path, establishing incremental growth rather than a single release.
Virginia Tech AI and Machine Learning Bootcamp — confirms $9,995 standard tuition for a 23-week part-time program, grounding the "paid bootcamps charge thousands" comparison in a first-party price.