Optimization · AI · Quantum
QuDecide turns complex optimization problems into practical, benchmarked software — combining strong classical methods with AI and quantum approaches where they genuinely add value.
Translate the real constraint system — objectives, limits, uncertainty — into something a solver can reason about.
Match algorithms to the structure of the problem rather than to the technology that happens to be fashionable.
Measure the result against credible baselines, under identical conditions, and report what actually moved.
What we do
Three connected capabilities. Most engagements draw on more than one, because a decision problem rarely respects the boundary between modelling, software and evaluation.
Mathematical models and hybrid solvers for scheduling, allocation, routing, search and other constraint-heavy decisions.
Reliable prototypes and reusable tools that connect machine learning with domain knowledge and measurable outcomes.
Honest evaluation of QAOA, quantum search and hybrid workflows against strong classical baselines — without assuming an advantage.
Work
Different industries, one recurring structure: too many valid options, hard constraints, and a real cost attached to choosing wrong.
Shifts, machines, maintenance windows, deliveries. Rules that cannot be broken sit alongside preferences that can be traded — and the schedule has to be regenerated whenever reality moves.
Finite resources against competing demands, where the interesting question is not only the optimum but how much the answer changes when an assumption turns out to be wrong.
Configuration, design and parameter spaces too large to enumerate, where structure in the problem can be exploited far more effectively than raw brute force.
Deciding quickly whether a candidate is admissible at all, so that expensive optimization is only ever spent on options that can actually be delivered.
A structured answer to “should we be looking at quantum for this?” — grounded in the problem’s structure and current hardware, not in the roadmap slide of a hardware vendor.
Client engagements are confidential. The categories above describe the classes of problem QuDecide is built to work on; specifics are discussed under NDA.
Our approach
Every engagement starts with the problem structure: objectives, constraints, uncertainty, and the cost of a wrong answer. We build a representative proof of concept, compare credible methods under the same conditions, and make the trade-offs visible.
That includes being willing to conclude that the simpler method wins. A benchmark that says so is still a useful result — usually a cheaper one.
We map the decision, its constraints and the baseline you already have. Deliverable: a written problem statement you can disagree with.
A representative instance is modelled and solved end to end — small enough to be fast, faithful enough to be meaningful.
Candidate methods are run under identical conditions and reported with their assumptions, limits and failure modes intact.
Code, documentation and a recommendation — including what we would not build, and what would have to change for that to shift.
About the founder
Milad Ghadimi is an optimization and quantum-algorithms researcher based in Dresden, Germany.
His work combines mathematical modelling, scientific computing, machine learning, and quantum algorithms. He focuses on converting complex research problems into reproducible software and evaluating new methods—including quantum and hybrid approaches—against credible classical baselines.
With experience spanning academic research and machine-learning engineering, Milad founded QuDecide to help organizations make difficult technical decisions based on measurable evidence rather than technological hype.
How we work
A new method is only interesting relative to the best straightforward alternative. That alternative gets implemented properly, not strawmanned.
Guarantees are stated as guarantees; empirical observations are stated as observations. The two are never blended into a speedup claim.
Results ship with the code, data and configuration needed to regenerate them. If it cannot be re-run, it is not yet a result.
“This does not pay off yet” is a legitimate outcome, delivered early rather than buried at the end of a long engagement.
Have a difficult decision problem?
Send a short description of the decision, the constraints around it, and how it is handled today. You will get a considered reply — not a pitch deck.