BottleTech Academy
The agent writes the code. You have to know whether it is right.
A seven-week, full-time internship for fresh software engineering graduates. 151 hours of structured work, 37 deliverables a mentor checks, and two scored checkpoints.
There is no application form on this site. BottleTech invites each cohort.
Review clinic · Week 4
async function fetchReport(id: string) { for (let attempt = 0; attempt < 3; attempt++) { const res = await fetch(`/api/reports/${id}`); if (res.status === 200) { return res.json(); } await sleep(200);// found in week 4: the retry never backs off } throw new Error('report unavailable');}- 7 weeks
- full time
- 148 items
- across 31 modules
- 37 deliverables
- each with a written acceptance check
- 151 hours
- of structured learning
- 2 checkpoints
- five criteria, scored 1 to 5
- 2–10 people
- per cohort
The program
Seven weeks, full time, inside a working software company.
You join a cohort of 2 to 10 people, work through the same weeks as everyone else, and hand work in to a mentor who either approves it or returns it with a comment.
Curriculum
Courses, readings, exercises and 37 deliverables, across 31 modules.
Day 0
A paid day 0 before week 1 sorts the machine, the shell and the stuck rule, so week 1 starts on time.
Hours
151 hours of that is structured learning. The rest of each week is live project work with a mentor.
Links
78 items link out to 70 external sources. Every link was checked by hand by 6 September 2026, and the platform re-checks all of them every night.
Cost
Paid resources are always optional and always paired with a free alternative in the same entry. An intern can complete the entire program at zero cost.
Acceptance
Every deliverable names what to submit and what a mentor will check. Nothing is graded on impression.
Live work
From week 5 you work real tickets from live BottleTech projects, not exercises.
One loop, in every track
Build it by hand. Direct the agent. Review what it wrote.
Five tracks run this loop: frontend, backend, database, DevOps and AI. Every one of the 148 items is marked AI-off, AI-on, or mixed, and the marking is not a suggestion.
AI-off
Fundamentals, by hand
Write it yourself first. An HTTP service with no framework, a command-line tool with no argument parsing library, SQL typed one query at a time, and a debugger instead of a guess.
A twenty-minute unaided live extension, run as one group session, checks the AI-off work at the end of weeks 1, 2 and 3.
AI-on
AI-native engineering
The same problems again with Claude Code, then the parts underneath: the messages loop, prompt caching, tool use, MCP servers, subagents, and eval sets with a documented pass bar.
In week 1 you rebuild your own morning's code with an agent, then write down what it got wrong.
Mixed
Judgment
Review, testing, evals, security and writing. A ranked review finding, a test that fails for the right reason, a PR description a stranger can approve without asking a question.
This is the track that decides both checkpoint scores that matter most.
Week by week
Structured hours fall as real work rises.
Week 2 is the heaviest at 29.25 hours of structured learning. Week 7 is the lightest at 13.75. The gap fills with live tickets and your capstone.
Week 1 · 25.25 h
Backend basics
Build it by hand in the morning, direct the agent in the afternoon
Week 2 · 29.25 h
Database and network
Know what is underneath before you automate it
Week 3 · 26.25 h
Frontend
Learn the cascade before you learn a shorthand for it
Week 4 · 26.25 h
AI agents and evals
Make the agent do real work on the real transport, then prove it works
Checkpoint
Week 5 · 14.75 h
DevOps and tickets
Structured learning drops; real project work takes over
Week 6 · 15.75 h
Capstone
Spec it, test it, build it, measure it, deploy it
Week 7 · 13.75 h
Ship and demo
Finish it properly and be able to explain it
Checkpoint
- 151 hours in total. Hours are estimates for one intern working at a normal pace.
- They exclude mentor time, stand-ups, and the project hours that fill out weeks 5 to 7.
- Week 1 opens on your cohort's start date. Each week after it opens seven days later, at midnight Amman time, and is due seven days after it opens.
Verification
A graduate who cannot check AI-generated code is a liability.
That sentence is the reason this program exists. Every week is arranged behind it.
The review clinic, every week from week 4
One hour. A mentor plants real defects in a pull request written by an agent. You review it and post your findings, ranked, in front of the cohort.
Four clinics: an agent's pull request in week 4, a second in week 5, a security review of your own capstone in week 6, and a peer's capstone in week 7.
Mentors rate the clinic the strongest signal they have of whether someone is ready for a real ticket queue.
AI-off means AI-off
No agent, no chat window, no inline completion beyond what the editor ships with. A twenty-minute unaided live extension, run as one group session, checks it at the end of weeks 1, 2 and 3.
Someone who cannot write a route handler unaided cannot review one.
AI-on does not mean trusted
Code that runs is not code that is right. Tests have to fail before they pass. Eval sets carry negative cases and Arabic cases that a naive prompt actually gets wrong.
Your week 1 critique has to name three concrete defects in agent output, each with a file and a line, plus one thing the agent did better than you.
Checkpoints
Scored twice, on the same five criteria.
Week 4 and week 7, with the same rubric both times, so movement is visible rather than remembered.
Checkpoint 1 — week 4: can you verify your own work
Ninety minutes with two mentors. The first ten minutes are with your laptop closed. Scores and written notes reach you the same day. A criterion under 3 sets what your mentor makes you practise in weeks 5 to 7.
Checkpoint 2 — week 7: are you ready for a real ticket queue
Two hours with two mentors, against your shipped capstone, your four review clinics, your merged tickets and your retrospective. The output is a written recommendation: hire, extend, or do not proceed.
| Criterion | What a 5 looks like |
|---|---|
| Understands the code they shipped | Walks a mentor through every line from memory, and can debug it live. |
| Writes clear specs and prompts | Writes a spec another intern could build from: files, scope, and a verification step. |
| Catches wrong AI output | Finds real defects the tests missed, ranks them by severity, and explains each failure. |
| Writes tests and evals | Tests fail for the right reason first, and the eval set catches real regressions. |
| Communicates well in PRs and demos | A reviewer who was not in the room approves without asking a question. |
- Every score cites a specific artefact you produced. A criterion scored without one is not scored.
- 4 or above on catching wrong AI output and on tests and evals means ready for a normal ticket queue, whatever the other three say.
- A 2 on either of those two means not ready, however strong the rest looks.
- You receive the written notes, and you have a right of reply.
Fit
Who this is for, and who it is not for.
This is for you if:
- You are a fresh software engineering graduate, or a final-year student close to finishing.
- You can write a small program on your own, even slowly, without help.
- You can work full time for seven weeks and hit a weekly deadline.
- You want to be corrected in writing, every week, by someone who actually read your code.
- You read and write Arabic, in any variety, and you can read technical English documentation. The linked courses and readings are in English; the code you write is bilingual Arabic and English.
- You have a GitHub account. BottleTech supplies the machine, so you do not need your own laptop.
This is not for you if:
- You want part-time or evening study. This is full time for seven weeks.
- You want to watch videos and collect a badge. 37 of the 148 items are deliverables a mentor inspects.
- You want AI switched on from day one. Weeks 1 and 2 have blocks where it is off, and that is checked.
- You are already a senior engineer. Week 1 starts at the shell, git, and a service with no framework.
- You want a guaranteed job. Checkpoint 2 produces a written recommendation, not an offer.
- You would rather go quiet than say you are stuck. That is the one failure this program cannot fix for you.
How to get in
Cohorts are invited, not applied to.
There is no open application form on this site today. BottleTech picks each cohort and sends an invite code.
If you have an invite code
- The link fills the code in for you. You add your name, your email, and a password of 12 characters or more.
- A cohort is 2 to 10 people. One code belongs to one cohort and stops working when that cohort closes.
If you do not have one yet
- Write to academy@bottletech.net.
- Send a CV, a link to code you wrote yourself, and two sentences about the hardest bug you have fixed.
- Start dates, whether the place is paid, and where you work from are confirmed per cohort. Ask in the same email.
For companies and universities
What this says about the engineers on your project.
- Clients ask how BottleTech trains people who ship with AI. This page is the answer, in public.
- Every intern is scored on whether they catch wrong AI output before it reaches a repository that matters.
- The full curriculum — every source, hour estimate and acceptance check, written for mentors — is available on request.
- Universities: if you teach software engineering and want the week-by-week plan, or want to send final-year students, write to us.
- Engineers: a cohort runs with one to three mentors. If you review code for a living and want to run a clinic, same address.
Questions
Questions people ask before they say yes.
Is it paid?
The learning material costs you nothing. Every paid resource is optional and carries a free alternative in the same entry, so an intern can complete the whole program at zero cost. Whether the place carries a stipend: BottleTech confirms this per cohort — ask at academy@bottletech.net before you accept.
Is it remote or in person?
The platform runs on Amman time. Each week opens at midnight Asia/Amman and is due seven days later, wherever you are. Whether it is remote or in person: BottleTech confirms this per cohort — ask at academy@bottletech.net before you accept.
Do I need to know AI tools already?
No. Week 1 opens with AI switched off, at the shell and git. Claude Code arrives on the afternoon of week 1, and the two courses that teach it are free.
How many hours a week is it?
It is full time. Structured learning is 25.25 hours in week 1, 29.25 in week 2, 26.25 in weeks 3 and 4, then 13.75 to 15.75 hours in weeks 5 to 7. The rest of those weeks goes to live tickets and your capstone.
What if I fall behind?
Say so early, in writing, to your mentor. A returned deliverable is normal; 32 of them exist so that work gets corrected. Your mentor can ask an admin to hold a week open for the whole cohort. There is no per-person extension in the platform today, which is why telling someone early matters.
What do I get at the end?
Merged pull requests in live repositories, a shipped capstone with a README and a runbook, written checkpoint notes against five criteria at week 4 and week 7, and a written recommendation from two mentors. Whether a certificate is issued: BottleTech confirms this per cohort.
Does it lead to a job at BottleTech?
Checkpoint 2 ends in one of three recommendations: hire, extend, or do not proceed. That is a recommendation, not an offer. Whether a role is open at the end: BottleTech confirms this per cohort, and will tell you before you start.
Can I apply? Where is the form?
There is no form today. BottleTech invites each cohort of 2 to 10 people and sends the code directly. Write to academy@bottletech.net with a CV and a link to your own code to be considered for the next one.
Do I need my own laptop?
No. BottleTech supplies the machine, imaged and checked in the paid day 0 the week before your cohort starts, so week 1 opens on one that already works.
Is any of this in Arabic?
The site is in English today, and the linked courses are in English. The code is bilingual: the week 1 service returns Arabic and English fields, and week 2 ends in a bilingual classifier. An Arabic version of this page is planned, with no date set.