Build AI features that survive production.
A six-week cohort for working engineers — context, evals, guardrails, agents, and the CI discipline that makes AI-assisted speed safe. Built and taught by Innoverse, from what we actually ship. Not vibe-coding.
AI makes it easy to build a demo and hard to ship something that holds. The gap between the two is engineering — evals, guardrails, observability, CI — and almost no one is teaching it.
This cohort is that missing discipline, taught for people who already know how to build software.
Who it's for
This is for you if
- You ship code for a living and want to add AI features that actually hold up.
- You've hit the wall where the demo works and production doesn't.
- You want the engineering discipline behind AI — evals, guardrails, CI — not another prompt-tips thread.
Not for you if
- You're brand new to programming — this assumes you already build software.
- You want a no-code or prompt-only shortcut.
- You want a credential to collect — ours is earned by shipping, not by showing up.
You leave with things that run.
A dataset and scorers that catch regressions before they reach a user — the thing most AI demos never have.
Input and output checks, injection handling, and graceful failure — so the feature degrades instead of embarrassing you.
You bring the problem — you'll ship one real AI feature into your own product or work codebase, wired end to end.
A pipeline where a bad prompt fails the build, the same way a broken test does.
Why join.
Free tuition
The founding cohort costs nothing — you just bring your own model API keys.
A free professional certificate
Finish and earn a professional certificate of completion — something real for your CV and LinkedIn.
Skills that outlast the hype
The production discipline — evals, guardrails, CI — that applies to every AI feature you'll ever ship, not one library.
Mentored by people who ship
Direct feedback from studio engineers who do this for a living, on your real work.
Something real to show
You leave with a production-grade feature already in your own codebase — portfolio-ready.
A path to the studio
Standout work can earn an invitation to paid work with the studio — a contract or a role.
Six weeks, six modules, one capstone.
LLMs from first principles
- Tokens and context windows, actually explained
- Choosing a model by task — not by hype
- Cost and latency as design constraints, not afterthoughts
Streaming, structured output & tool calling
- Streaming text and typed structured output end to end
- Tool / function calling and MCP servers
- Rendering and persisting message parts
Agents vs. workflows
- When a loop beats an orchestrated workflow — and when it doesn't
- Task decomposition and the evaluator–optimizer pattern
- Routing simple work to cheaper, faster models
Eval-driven development
- Evals as the unit tests of AI — build a real dataset
- LLM-as-a-judge plus deterministic scorers
- Catch regressions before your users do
Guardrails & reliability
- Input/output guardrails, injection and refusal handling
- Retries and graceful failure paths
- A cheap guardrail model in front of the expensive one
Observability, CI & shipping
- Tracing, prompt caching, and cost/latency budgets
- Evals wired into CI so a bad prompt fails the build
- “If it can't survive production, it isn't done.”
Ship a feature into your own codebase.
Bring a real problem from your product or your work. You'll ship one production-grade AI feature — eval suite, guardrails, tracing, and evals gating a CI pipeline — reviewed by studio engineers in the final office hours. Ship it and you earn your certificate. You leave with something already in your codebase, not a throwaway demo.
Draft outline — the syllabus is being finalized and may shift before the first cohort.
From the waitlist to shipped.
- 01Join the waitlist
Leave your email — you'll be first to hear when a cohort opens, with dates and how to apply.
- 02Apply when it opens
A short application. We take people who already do the work and want to get sharper — not beginners.
- 03Get into the cohort
A small group, closely mentored, so every project gets real attention.
- 04Six weeks, building
Weekly lessons you do on your own time, plus live office hours with the Innoverse team.
- 05Ship your capstone
A production-grade project in your own codebase — reviewed by Innoverse engineers.
- 06The standouts → the studio
Do exceptional work and you may be invited to keep building with the studio — paid client work or a role. Never promised, always earned.
Innoverse — a product engineering studio in Belgrade building web, mobile, and AI products for founders across the EU and US. The same engineers who ship these systems for clients teach the program — from the real thing, not slides.
This is a founding cohort: new, small, closely mentored — that's the point, not an apology. We'd rather run one excellent room than a crowded one.
See the studio →Before you apply.
Do I need to be an AI expert already?
No — you need to be a working engineer. We start from first principles and move fast because you already know how to build software.
What are the prerequisites?
Comfort with TypeScript and a JavaScript framework (React/Next). You'll bring your own model API keys (OpenAI/Anthropic) — tuition itself is free.
How much time does it take?
About 5–6 hours a week, built to fit around a full-time job: async lessons you do on your own schedule, plus a couple of live office hours (recorded if you miss them).
Is it remote?
Yes — fully remote and global. Lessons are async; office hours run live over video and are recorded.
What does it cost?
The founding cohort is free. We're comping the first run to get it right and build the first case studies — all we ask is that you commit to finishing and give honest feedback. Later cohorts will be paid.
Does the capstone have to use my employer's codebase?
No — any real codebase you have the rights to work in. A side project works just as well; what matters is that it's a production-grade feature in real code, not a toy exercise.
Do I get a certificate?
Yes — finish the cohort and you earn a free professional certificate of completion, something real for your CV and LinkedIn. It's earned by shipping the capstone, not by attendance.
Why trust a brand-new academy?
Because it's run by an active product studio — you're taught from real shipped work, and the first cohort is deliberately small and closely mentored. We have no graduates yet, and we won't pretend otherwise.
Could this lead to working with the studio?
Standouts may be invited into paid work with the studio — a contract or a role. It's earned through the work, never promised up front.
Get in the first cohort.
The founding cohort starts mid-September, and it's free. Join the waitlist and you'll hear the moment applications open — with exact dates and how to apply.
No spam — just one email when a cohort opens.