Engineering Coaching
AI gets you to 80%. The last 20% is yours to own.
You built something with AI, and it worked, right up until the project hit real complexity. Infra, edge cases, false assumptions, and the architecture and judgment calls are the last 20%, and that's where developers reach out to me.
I coach you through exactly those, turning a stalling proof-of-concept into an engineered system you can ship, test, and defend. On your own real project, over 6 to 12 weeks, with every pull request deeply reviewed by me.
Most work with me 1:1, but there are also small-group cohorts (Agentic AI, Rust) and team training.
Prefer to talk first? Book a free call or tell me about your project.
Is this you?
Wherever you're starting, we begin from exactly there. Already senior and just want to go deeper on advanced Python, Rust, or production AI? That works too, and the sharpest next move is usually the Rust cohort.
1:1 coaching on your own project
This is where I spend most of my time. You bring a real project, new or existing, and we make it something you can change safely, test, and defend. Typically a full-stack, layered, tested Python app, and I prefer you come with a challenging idea of your own which we scope out together.
- Weekly 1:1 coaching calls on the problem you're actually stuck on
- Every pull request deeply reviewed by me: code quality, maintainability, architecture, performance
- Your own real project, taken from proof-of-concept toward something you can ship
- 6 to 12 weeks. You finish owning the tech, with the confidence to present it to employers and tech leaders
No project in mind yet? Start from a fixed-scope six-week template (Snipster or a Django SaaS).
"I was building a full-stack app mostly through AI-assisted coding and needed guidance to take it from proof of concept to production. Bob coached me through the parts I couldn't validate on my own: Django settings architecture, payment security, error handling, query optimization. More than anything, his coaching gave me the confidence to actually ship."
From where they started to what they shipped
Developers I've coached have shipped a legal citation generator for Canadian courts, a distributed lock over gRPC, an AI bookmark CLI, a published pytest plugin, and a Django payroll SaaS.
See the projects → · More stories →
Why this works when AI writes the code
The fastest way I grew was getting my code reviewed by people more experienced than me. That's quietly become rare: most review is generated now, and generated review tells you what a pattern is called, not whether it was the right call in your codebase.
Every pull request in my coaching is deeply reviewed by me. AI raised the baseline for everyone; what it amplifies is judgment: system design, architecture, code that stays maintainable at scale. That's best learned from someone who's been through it many times.
I bring 20 years to it: from VBA Excel automation in finance to 10+ years at Sun/Oracle doing support tool and construction payment system development, co-delivering Talk Python trainings, co-founding Pybites building two coding platforms, and having coached over 150 developers.
The three pillars of modern software engineering
Whatever we build, the coaching rests on the same three pillars: what turns AI output into software you can change safely and defend.
Decision-making
Software has always been about the choices, not the typing: what to build, what to skip, what breaks later. AI makes the typing cheap and the choices matter more.
The control layer
The model is the easy part. The engineering is the layer around it: the patterns that make a nondeterministic tool behave the same way every time.
Design over code
The productivity is in the questions you ask before coding. One student's 1,069-line AI app became 156 lines once the design was right.
Working with AI, deliberately
I'm not going to ask you to use AI less. You'll use it throughout, at whatever speed you want, and we work on the part that decides whether that speed compounds or costs you later:
- Reading a diff you didn't write and deciding quickly whether it earns its place
- Knowing when to accept, when to redirect, and when to throw it out and write it yourself
- Keeping your tests honest, so you don't end up with generated code confirming generated code
- Spotting output that's plausible rather than correct, the failure mode that costs the most later
This is hard to pick up from a course or a book, because it only shows up on real code with real stakes. It's most of what my PR reviews are actually about.
The thinking behind my approach
Why a coach, when the answers are free
AI hands you answers for free. What it can't do is tell you whether the answer is right, show you where you are, or make you finish. That work got more valuable when answers got cheap, not less.
- A map, not more answers. Someone who has shipped this before tells you where you are and what the next move is, so you stop drowning in plausible options and start making progress.
- Calibrated to you. An answer is the same for everyone who types the question. Coaching meets you where you are, whether that's a first real project to production or going deeper as a senior.
- Accountability that makes you finish. A person expecting your next pull request moves you in a way a chat window never will. Most of what stalls a project isn't the code, it's finishing it.
Also available: cohorts and team training
Beyond 1:1, I co-run two hands-on 6-week cohorts, each pitched at a level. Same format: weekly calls, detailed PR reviews, and a small group building the same thing.
Agentic AI
For intermediate to advanced Pythonistas.
Getting an agent to work once is a demo. Making it do the same thing every time, tested and deployable, is the engineering. You build a predictable system around a nondeterministic model, not flaky POC glue: function calling, three interfaces, 95%+ test coverage, Docker deploy. Co-led with Juanjo Expósito.
Scripter to Rust
For advanced Pythonistas ready for systems work.
The strongest next move once Python starts costing you on performance and safety. Learn where scripting languages hit their limits and think like a systems engineer: a hands-on Rust + PyO3 project, benchmarked against CPython. Co-led with Jim Hodapp.
Coaching for teams
Your team doesn't need more generated code, it needs better feedback loops. Speed without engineering review compounds into technical debt. I run the same approach with engineering teams: your developers build something real together, get deep code review, and carry those habits back to your codebase. Private cohorts (Python, Rust, or Agentic AI) or hands-on coaching against your own repo.
Frequently asked questions
What does it cost? The Agentic AI and Rust cohorts are €2,000 each. The fixed-scope 1:1 Python program is €1,500. Custom 1:1 on your own project varies, so I price it after we talk, in a written proposal you see before committing. Most developers recoup the fee inside one role move, promotion, or paying client.
How long does it last? A focused 6-12 week program of weekly coaching. Some stay on longer as their projects grow.
Can this help with the job market, not just my code? Directly. Shipped work you can defend in a review and talk about in public is the proof that opens doors, and reads as senior in interviews. Piotr made Senior AI Engineer, Heather moved from Excel/VBA into software, Rodrigo went from chemical engineering to a remote dev role.
Is coaching still worth it when AI writes the code? That's exactly why. When code is cheap to generate, producing it stops being worth much on its own; the value moves up to architecture, review, and knowing when generated code will hold in production. That's what the tools amplify, and where a coached developer pulls ahead. Josh built a Rust JSON parser that beat CPython's C stdlib and wrote about why learning deeply matters more now, not less.