AI & Agent Education · Live Cohorts & In-House Training

AI won't take the job.
Someone who understands it will.

LeanProgressIQ teaches working professionals to use, apply and verify AI and autonomous agents inside their actual jobs — and to keep up as the requirements of those jobs keep moving.

"I don't try. I verify." — Every stage ends with evidence you produced yourself, not a certificate you were handed.
Begin the Six-Stage Ascent Talk to the Architect
Scroll to Climb the Curriculum
STAGE 01 · 1,940M (GHANDRUK RIDGE)
FOUNDATIONS · NO PRIOR TECH REQUIRED

AI Literacy — What These Systems Actually Do

Models, tokens, context, limits · Why AI is confidently wrong

Most AI training starts with tricks. This starts with mechanics. You learn what a language model is actually doing when it answers you, why it invents facts, what a context window costs you, and where the real boundary sits between "impressive demo" and "safe to use at work". By the end you can predict when a tool will fail before you rely on it.

4 Weeks
Typical Duration
Zero
Coding Prerequisites
Any Role
Entry Point
How Models Work Hallucination & Why Context & Memory Tool Landscape Data & Privacy Basics
STAGE 02 · 2,600M (CHOMRONG GATEWAY)
INSTRUCTION & VERIFICATION

Getting Output You Can Actually Defend

Specification over prompting · Checking before shipping

"Prompt engineering" is mostly specification writing — the same skill that makes a good requirement, a good brief, or a good work instruction. You learn to state the task, the constraints, the format and the acceptance criteria, then to verify the answer against evidence rather than against how confident it sounds. This is the stage that separates people who use AI from people who trust it blindly.

Specify
Not Just Prompt
Verify
Against Evidence
Rubrics
You Write Yourself
Task Specification Acceptance Criteria Source Checking Rubric Authoring Red Teaming Your Own Output
STAGE 03 · 3,230M (DEURALI GLACIER GORGE)
LEAN APPLIED TO AI ADOPTION

AI in the Workflow — Lean on Your Real Process

PDCA · 5S · Waste elimination · Measured before and after

This is where LeanProgressIQ differs from a tools course. You bring a real process from your own job — a report, a handover, a planning cycle, a support queue — map it, find the waste, and apply AI only where it removes waste rather than where it looks impressive. Twenty-five years of Lean practice in Tier-1 manufacturing and ERP rollouts, pointed at the way your work runs today.

PDCA
Improvement Loop
Your Process
Not a Toy Case
Before/After
Measured Cycle Time
Process Mapping Root Cause Analysis Waste Elimination PDCA & 5S ERP & Ops Context
STAGE 04 · 3,700M (MACHAPUCHARE BASECAMP)
AGENTS, TOOLS & SAFE AUTONOMY

Agents at Work — Tools, Autonomy and Limits

What an agent loop is · Where to give it hands, and where not to

An agent is not a smarter chatbot — it is a loop with tools, permissions and consequences. You learn the anatomy of that loop from a framework built in the open: tool calling, permission gates, human confirmation checkpoints, multi-agent delegation, and the failure modes each one introduces. Taught on aura-code, the MIT-licensed agent framework created and architected in-house.

25+
Tools Dissected
30
LLM Providers Compared
Live Loop
Run in the Session
Agent Loops Tool Calling Permission Models Multi-Agent Delegation MCP & Integrations
STAGE 05 · 4,130M (ANNAPURNA SANCTUARY)
EVALUATION & EVIDENCE

Evaluation & Trust — Proving It Actually Works

Benchmarks, rubrics, longitudinal testing · Evidence over vibes

Organisations adopt AI on anecdote and abandon it on anecdote. This stage teaches measurement: how to build a benchmark for your own domain, write scoring rubrics, run the same suite over time, and read the result honestly. Built on the same practice behind a 320-question evaluation harness and a public 50-day longitudinal agent trial.

320-Q
Reference Benchmark
50 Days
Public Trial Data
Your Domain
Own Eval Suite Built
Benchmark Authoring Scoring Rubrics Preference Ranking Longitudinal Testing Red Teaming
STAGE 06 · 6,993M (APEX)
CAPSTONE

Build & Stay Current — Your Own System, Kept Current

Ship something real · Then keep it alive as the field moves

The capstone: you build and deploy a working AI or agent system for your own role, with its own evaluation suite, and you leave with a practice for staying current — how to read a release note, test a new model against your own benchmark, and decide whether it changes anything. Skills expire fast in this field; the method for renewing them does not.

1 System
Deployed & Yours
1 Eval
Suite That Outlives It
Ongoing
Currency Practice
Capstone Build Deployment Model Migration Continuous Evaluation Career Currency
The Ecosystem · Every System Taught Here Was Built Here

Projects Behind the Curriculum

LeanProgressIQ does not teach from slides about other people's tools. Every project below was designed, built and verified in-house by Dušan Milosavljević — Creator & Systems Architect, and each one feeds a stage of the curriculum.

🏔️

leanproiq.com

Home of LeanProgressIQ

The main LeanProgressIQ domain — training programmes, Lean consulting and AI adoption work for teams and individuals.

Creator & Architect · Dušan Milosavljević
📐

Lean Progress IQ

Lean Practice Platform

The Lean side of the house: PDCA, 5S, E-Kanban and waste elimination methods drawn from 25 years in Tier-1 manufacturing and SAP rollouts.

Creator & Architect · Dušan Milosavljević
⚡

aura-code

Core Agent Framework

Autonomous agent loop with 30 LLM providers, 25+ tools, multi-agent orchestration and 1,559 passing tests. The teaching platform for Stage 04.

Creator & Architect · Dušan Milosavljević
📱

aura-droid

Native Android Client

Kotlin Android client running the full agent loop on mobile over high-performance WebSocket transports.

Creator & Architect · Dušan Milosavljević
🎯

Aura Benchmark & 50-Day Trial

Evaluation Harness

320-question agent evaluation harness plus a public 50-day longitudinal trial. The evidence base for Stage 05.

Creator & Architect · Dušan Milosavljević
🚀

enables

Provider Bridge

Lets the Claude Code CLI run against almost any upstream AI model provider — the practical lab for comparing models in Stage 04.

Creator & Architect · Dušan Milosavljević
🕸️

agentmesh

Visual Workflow Builder

Node-based canvas for assembling, configuring and executing multi-agent workflows — how non-programmers build agents in Stage 06.

Creator & Architect · Dušan Milosavljević
🛡️

aura-pulse & Clients

Tauri / Rust / Telegram

aura-mathetes (Tauri/Rust desktop), aura-pulse (telemetry & benchmarks) and lakaura-bot (Telegram voice agent).

Creator & Architect · Dušan Milosavljević
⚖️

Serbian Legal LLM

Fine-Tuning Pipeline

~1,600 statutory acts and 11M+ cleaned words into Vertex AI supervised fine-tuning, plus Granite and Gemma small-model pipelines.

Vertex AI · Granite · Gemma
Creator & Architect · Dušan Milosavljević
🗺️

Alpine Portfolio

WebGL Terrain Site

The scroll-driven Himalayan terrain engine behind this very page — real DEM heightmap data rendered live in Three.js.

Creator & Architect · Dušan Milosavljević
leanprogressiq — live cohort session
How the Teaching Works

Lean Method, Applied to AI Adoption

The same discipline that removes waste from a production line removes waste from AI adoption: understand the system, find the real constraint, change one thing, measure it.

🔍

Mechanics Before Tricks

You cannot judge a tool you do not understand. Every stage starts with how the thing actually works, so your judgement survives the next model release.

🧪

Evidence Over Vibes

"It felt better" is not a result. Learners build rubrics and benchmarks for their own domain and report measured before-and-after, not impressions.

♻️

PDCA on Real Work

Plan-Do-Check-Act against a process you actually own. No sandbox exercises that evaporate the day the course ends.

🚧

Waste, Not Novelty

AI goes where it removes waiting, rework, over-processing or handover loss — not where it produces the most impressive demo for management.

🛡️

Safe Autonomy

Agents get hands only behind permission gates and confirmation checkpoints. Learners are taught where to stop delegating, and why.

🧭

Currency as a Practice

The field moves monthly. You leave with a repeatable routine for testing what is new against your own benchmark instead of chasing announcements.

New Work Requirements · Who This Is For

Staying Employable While the Job Description Changes

Job descriptions are being rewritten around AI faster than training catches up. These are the roles LeanProgressIQ works with, and what changes for each.

🏭

Operations & Manufacturing

Planners · Supervisors · Lean Teams

Planning reports, master-data checks, shift handovers and root-cause write-ups — the paperwork layer around the line, where AI pays off first. Taught with real SAP and E-Kanban context.

🏢

Enterprise & ERP Teams

SAP · Logistics · Back Office

Transaction validation, migration checks, documentation and support triage. Taught by someone who ran SAP master data and greenfield rollouts, not by someone reading the brochure.

💼

Professionals & Managers

Analysts · PMs · Consultants

Research, drafting, analysis and reporting — plus the harder skill of defending an AI-assisted conclusion in a room full of people who will challenge it.

🎓

Educators & Trainers

Teachers · L&D · Coaches

Assessment design that survives AI, materials generation, and teaching AI use honestly. Grounded in 11+ years of B2B executive coaching and TEFL-certified pedagogy.

⌨️

Developers & Technical Staff

Engineers · IT · Automation

Agent architecture, tool design, permission models, evaluation harnesses and model routing — taught from the source of a framework with 1,559 passing tests.

🧑‍🤝‍🧑

Teams & In-House Cohorts

Private Programmes

The full six-stage ascent run inside one organisation, on that organisation's own processes, with a shared evaluation suite left behind as the deliverable.

Track Record · 2001 to Present

Where the Teaching Comes From

"Not a series of career changes — twenty-five years spent studying and improving different kinds of systems."

2024 – PRESENT
Creator & Systems Architect — aura-code
LeanProgressIQ · Published on npm

Designed a full autonomous agent loop with streaming and confirmation checkpoints, 30 LLM providers with runtime switching, 25+ integrated tools, and 1,559 passing tests — now the teaching platform for Stages 04–06.

2024 – PRESENT
AI Evaluation & Training Specialist
Aura Benchmark · 50-Day Longitudinal Trial

Authored a 320-question benchmark with rubrics and execution sandboxes, and ran a public 50-day longitudinal agent trial with automated daily reporting — the practice taught in Stage 05.

2015 – PRESENT
Executive Trainer & B2B Coach
Learnlight · Bibo Global · TEFL Advanced (120h)

Eleven-plus years delivering high-stakes B2B training to corporate executives — presentations, negotiations and correspondence — with assessment rubrics and pedagogical design behind it.

2013 – 2020
Lean & Enterprise Systems Practitioner
Yazaki · Draexlmaier · SAP ERP

SAP superuser work on master data and logistics execution, a 2017 greenfield investment rollout, and E-Kanban with barcode scanning deployed across Romania and Serbia — the Lean material behind Stage 03.

2001 – 2013
Software Engineer & Agency Systems Superuser
University of Novi Sad (FTN) · Milotours · AMADEUS & Galileo

Trained as a Software Engineer (Informacioni Inženjer) at FTN in 2001, then twelve years running agency IT, networks and GDS connectivity. Lufthansa Experts Sales Award (2007); "Friend of Malaysia" (2005).

Creator & Architect of Everything on This Page

"I design, direct and verify intelligent systems rather than write their production code — and I teach exactly the way I work."

Dušan Milosavljević is the founder of LeanProgressIQ and the creator and systems architect of every project listed on this site — aura-code, aura-droid, the Aura Benchmark and 50-day trial, enables, agentmesh, aura-pulse and its clients, the Serbian Legal LLM pipelines, leanproiq.com, Lean Progress IQ and the WebGL terrain engine rendering behind this page.

Trained as a Software Engineer (Informacioni Inženjer) at the University of Novi Sad (FTN), he then spent twenty-five years inside real systems: travel platforms and agency IT, SAP ERP rollouts, logistics, and Lean manufacturing at Tier-1 automotive plants. Through Lean — PDCA and 5S — he learned to look past symptoms to root causes. Alongside it runs language: C2 English and eleven years teaching professionals, which is where the judgement about the quality of what is said was built. Today that judgement is pointed at AI.

That combination is the reason this curriculum exists. AI training is usually written either by engineers who have never run an operation, or by trainers who have never built a system. LeanProgressIQ is written by someone who has done both — and every claim taught in a session has a repository, a benchmark or a deployed site standing behind it.

PHASE 01
Deconstruct Mechanics
Reverse-engineer the information flows and dependencies before proposing a single change.
PHASE 02
Find the Real Waste
Separate the symptom everybody complains about from the constraint that actually causes it.
PHASE 03
Direct the System
Specify, delegate and gate the work — to people and to agents — with explicit acceptance criteria.
PHASE 04
Verify, Then Claim
Nothing is called finished until a test, a benchmark or a measurement says so.
🎓
Software Engineer (FTN, 2001) University of Novi Sad — Fakultet tehničkih nauka
🗣️
C2 English · TEFL Advanced (120h) 11+ years B2B executive coaching
⚙️
SAP ERP Superuser · Lean Practitioner PDCA · 5S · E-Kanban · Greenfield rollout
Basecamp · Enrolment & Enquiries

Start the Ascent

Individual cohorts, private in-house programmes, and AI adoption consulting for teams — plus Lean process optimisation and enterprise ERP work.

Enrolment Email leanproiq@gmail.com Main Site leanproiq.com Open Source Ecosystem github.com/DusanCar-sudo
Locations & Phones +84 37 612 5323 · +381 63 7692687
Architect's CV (PDF)