PixioLab

PixioLab AI Mentor

Become an AI engineer in 45 days. One step at a time.

A day-by-day path for software engineers. Build one real system in your own GitHub, prove it works with numbers, and have a mentor review your code at every checkpoint.

Week 1 and the roadmap guide are free. No card needed.

pixiolab.com/path
The PixioLab path dashboard on Day 18, with today's step and the 45-day grid
days, one step each
45
days, one step each
checkpoints with repo checks
7
checkpoints with repo checks
guides, 697 pages
10
guides, 697 pages
knowledge-check questions
36
knowledge-check questions
weeks of interviews and job search
4
weeks of interviews and job search
An engineer working on a laptop in a quiet, dark office
Today · Day 18 · 1–1.5 h
Hybrid search
1 Build a BM25 index over your chunks
2 Fuse it with vector search using RRF
3 Unit-test the fusion

How it works

One thing to do today, and a way to prove you did it

Most courses hand you hundreds of pages and leave you to it. The path tells you exactly what to do next, and each week ends with evidence in your own repository.

  1. 01

    Open today's step

    One task, about an hour on a weekday, with numbered instructions, a clear finish line and the mistake most people make.

  2. 02

    Build it in your own repo

    Every day adds to one real system, Sourcebook, in your GitHub. You tag a release at the end of each week.

  3. 03

    Pass the checkpoint

    Each week ends with criteria you tick, a short knowledge check, and automatic checks of the repo itself.

  4. 04

    Get your code reviewed

    On the Mentor plan, a reviewer reads each checkpoint line by line and approves it or tells you what to fix.

Inside a day

Instructions you can follow at 9 pm on a weeknight

Every one of the 45 days has numbered steps, a “done when” test, the common mistake to avoid, the chapter to read and the companion code to compare against. Students get extra work on the same day.

  • Real file names and commands, not vague advice
  • Weekend days build exactly what the checkpoint asks for
  • A six-question knowledge check closes every week
pixiolab.com/path/d18
Day 18 in the app: numbered steps, done-when and common-mistake boxes, and the week's resources

What you build

You finish with evidence, not a certificate

Over six weeks you build Sourcebook, a cited-answer assistant over a public document set, then Groundtruth to measure it and Tripwire to stop regressions in CI. By Day 45 your README opens with a results table like this one, every number measured by you.

  • Hybrid retrieval and checked citations
  • A bounded agent mode with traces
  • An evaluation harness and judge
  • Injection defenses and a CI gate
README.md · Resultsn = 50 questions
recall@10, vector only → hybrid0.71 → 0.86
Faithfulness pass rate (judge)88% → 95%
Attack success rate, 24 attacks29% → 4%
p95 latency · cost per 100 queries2.4 s · $0.41

Illustrative numbers. Yours come from your own golden dataset.

The library

Ten guides behind every step

697 pages of explanations, diagrams, tested code and job-search playbooks. The path tells you which chapter to read, and when.

Companion code on GitHub →
Guide 00: The 45-Day AI Engineer Roadmap
Guide 01: RAG, Explained
Guide 02: Agentic Design Patterns
Guide 03: Evaluation Field Guide
Guide 04: Production AI Features
Guide 05: The Portfolio Project Catalog
Guide 06: System Design Interviews
Guide 07: AI Engineering Interview Q&A
Guide 08: When You Don't Know the Answer
Guide 09: Get the Offer

Look inside

Diagrams that explain one idea each, code you can run, worked examples and self-checks. Hover the pages to spread them out.

Browse the guides
Guide 01 page: hybrid search diagram and a worked example
Guide 02 page: runtime architecture for deploying agents
Guide 03 page: debugging a failed agent trace step by step
Guide 04 page: production code with explanations
A mentor pointing at a learner's laptop while explaining a change
Week 3 checkpoint · v0.3Changes requested
  • ✓Repository is public
  • ✓Tag v0.3 exists
  • ✓data/README.md is in the repo
  • ✓No .env committed
To change before this passes
Out-of-scope questions still get an answer. Add the abstain rule from Day 19 and show the three refusals in the README.

The Mentor plan

A person reads your code every week

Tag your release and submit it from the checkpoint page. The site checks the repo first, so a reviewer only spends time on work that's ready. Then they read it against the week's criteria, comment on GitHub line by line, and approve it or tell you exactly what to change.

  • A decision within 5 business days
  • Resubmit until it meets the bar
  • A public profile of approved checkpoints, linked to your code

The plan

Six weeks, a final stretch, then interviews

Each week ends with a tagged release of your project. Week 1 is free, so you can see how it feels.

  1. Week 1Free
    How LLMs work, and your first call
    v0.1
  2. Week 2
    Prompts, schemas and tools
    v0.2
  3. Week 3
    Retrieval and RAG: Sourcebook v0.3
    v0.3
  4. Week 4
    Agents and tools: a bounded loop
    v0.4
  5. Week 5
    Evaluation and production: Groundtruth
    v0.5
  6. Week 6
    Safety, ML foundations and the ship gate
    v0.6
  7. Final
    Polish, publish, apply
    v1.0

After Day 45, the full program continues with a four-week track through system design, technical Q&A, questions you don't know, and the job search itself: referrals, take-home assignments and negotiating the offer.

Pick how much help you want

Start free. Upgrade when Week 1 has earned it.

Questions

Who is this for?

Software engineers who can write Python, use Git and read a stack trace, and want to move into AI engineering. No machine-learning background is needed.

How much time does it take?

On the working-engineer track, 10 to 12 hours a week: about an hour each weekday and longer blocks at the weekend. The student track is 20 to 30 hours a week. If it takes you eight or nine weeks, that's normal.

What else will I pay for?

Model API usage, usually $10 to $40 over the 45 days if you develop on a small model. Everything else the path uses is free or open source, and there's a zero-budget route with local models.

Do I have to follow the order?

The path always points you to your next step, and each week builds on the last, so the order matters. You can still open any step you have access to, and skip what you already know.

What happens in a review?

You tag the week's release, the site checks the repo automatically, and you explain one decision and one fix in your own words. A reviewer reads the code against that week's checklist and sends written feedback that points to files and lines, with a decision: approved, or changes to make first.

Will this get me a job?

It's built for that, but no course can promise one. You finish with a public repo with measured results, four weeks of interview practice on your own and with a peer, and a job-search guide that covers target roles, referrals, take-home assignments and negotiation. The interviews are still yours to win.

Is it a subscription?

No. The full program is one payment and it's yours to keep, including updates to the guides. The mentored program adds six months of reviews from a person, by application, and includes the full program for good.

Can I start free and upgrade later?

Yes. Week 1 is free, and your progress carries over when you buy the program. If you own the program and later want a mentor, its price is credited.

Day 1 takes about an hour.

Set up your repo, see the whole 45-day plan, and decide if the path fits you. It's free.

Start Week 1 free