Take the guesswork out of the decision.
Material, supplier, site location, design parameter: once the options multiply, intuition runs out. OptamLab brings design of experiments, surrogate modeling, multi-objective optimization and multi-criteria decision analysis into a single verifiable workflow.
Enterprise demo · academic collaboration · early access
Real screen: a lightweight material selection case for a rail vehicle, taken from a peer-reviewed publication and re-solved by the engine.
The method is not your real job, yet it takes your whole day.
A serious selection decision is spread across four separate tools today: design of experiments in one place, regression in another, the decision matrix in a spreadsheet, the report in a word processor. Every copy and paste in between is a chance for error, and every assumption goes unrecorded.
How it works today
- Design of experiments, model, decision and report sit in separate tools; the gaps are bridged by hand.
- Method and normalization choices are never written down, and six months later nobody remembers them.
- Reproducing the same result from the same data is usually not possible.
- When a reviewer or a client asks "why this alternative?", there is no trail to defend it.
With OptamLab
- One flow from definition to report; the data is entered once.
- Every method and variant choice is recorded and stated plainly in the report.
- Every run carries a configuration hash: same input, same numbers.
- Sensitivity, rank reversal and Monte Carlo robustness analyses come built in.
What is needed, as much as needed.
This is not a slogan, it is how the product works. The most information from the fewest experiments, a defensible result from the fewest runs.
Fewer runs
Orthogonal arrays measure the effect of every parameter with a fraction of the runs a full factorial plan would demand.
Spend the budget where it counts
A surrogate model removes the need to rerun an expensive solver for every candidate design; the search happens on the model.
The report is not written from scratch
Methods and results arrive as a draft, with the variants used declared. Your time goes into checking, not typing.
Still defensible in six months
Every run carries a configuration hash. The answer to "why this alternative?" lives in the file, not in someone's memory.
Four analysis modules, one data model.
The modules feed one another: the Taguchi experiment matrix flows into optimization, the Pareto front into decision analysis, and every result into the report layer. Because they work on the same data model, nothing has to be converted or carried across by hand.
Multi-criteria decision analysis
Scores and ranks alternatives against defined criteria and weights, then tells you how robust that ranking is. Weights can be derived from the data or supplied by expert judgment.
- 18 ranking methods and 12 weighting schemes
- Sensitivity, rank reversal and Monte Carlo robustness checks
- Fuzzy variants for uncertain inputs
- Where the literature holds more than one convention for a method, the choice is left to the user
Fuzzy MCDM
Inputs are not always exact numbers. When an expert says "about 40 kN" or "high", rounding that to a single value destroys information. The fuzzy module carries the uncertainty along with the data and produces the ranking under that uncertainty.
- Membership types: interval, triangular (TFN) and trapezoidal numbers
- Fuzzy TOPSIS, VIKOR, COPRAS, OCRA and ARAS
- Fuzzy AHP weighting with Buckley's method and linguistic scales
- Probability-based ranking and group decision aggregation
Taguchi design of experiments and ANOVA
Measures the effect of each parameter with a minimum number of experiments, finds the optimum level combination and tells you whether it is statistically significant.
- 8 standard orthogonal arrays (L4–L27) with automatic array selection
- 3 signal-to-noise types, dynamic S/N and multi-response analysis
- ANOVA: F value, p value, percentage contribution, error pooling
- Assumption tests and a verdict on the confirmation run
Surrogate modeling and multi-objective optimization
Builds a model from data in place of expensive simulation, searches over that model, and derives the trade-off front for objectives that conflict with one another. Which point to take from the front is settled by method as well.
- DOE: Latin hypercube, Halton, Hammersley, factorial, Box–Behnken, central composite
- Surrogate models: response surface, radial basis function, neural network, automatic selection by cross-validation
- Optimization: Nelder–Mead, differential evolution, NSGA-II, MOPSO, hybrid memetic
- Pareto front, hypervolume, convergence diagnostics and decision-supported point selection
Report layer
When the analysis is finished, the methods and results sections are produced as a draft, together with an explicit statement of the variants used. AI writes the prose; the engine produces the numbers, and every number in the report is checked against the analysis output.
- Methods, results and discussion sections; tables and formulas
- DOI-verified citations and method references
- Word and LaTeX output; Turkish and English
- Every setting that departs from the default is noted in the report
We do not assert. We leave it checkable.
A methods application is worth exactly what its arithmetic is worth. OptamLab's engines are validated against peer-reviewed publications and open reference implementations; where results diverge, that is flagged rather than hidden.
| What | How it is validated | Status |
|---|---|---|
| Case validation | Worked examples from peer-reviewed publications are re-solved by the engine using their inputs verbatim, then compared against the published values. The cases can be inspected without an account. | 108 cases 83 decision · 20 Taguchi · 5 optimization |
| Method cross-check | Ranking methods are compared against pymcdm, an open-source reference implementation, on identical inputs. | 18 methods |
| Reproducibility | Every run carries a configuration hash covering the data, weights, method and parameter choices. | On every analysis |
| Convention honesty | Where the literature defines a method in more than one way, the choice is exposed to the user and the one applied is declared in the report. | Declared in the report |
| What is out of scope | Methods that are not yet on the production path are marked "coming soon"; they are not presented as if they existed. | Marked plainly |
Your data stays yours.
R&D data is often a company's most sensitive asset. The infrastructure was built on that assumption.
EU data residency
Data is hosted in the Frankfurt region and processed under KVKK and GDPR.
Tenant separation
Row-level security policies separate each account's data at the database level.
Export and deletion
You can export your analyses at any time, and view or delete the data held in your account.
Immutable record
Saved analyses are locked; editing creates a new version and the previous one is preserved.
Teams that have to decide.
R&D centers
Justified selection among material, process and design alternatives; efficient use of an expensive simulation budget.
Manufacturing and quality
Improving process parameters with few experiments, managing tolerance and variation.
Academia
Work that demands methodological accuracy and reproducibility; a publication-ready methods section.
Consulting
Delivering a defensible, traceable and repeatable decision file to the client.
OPTAM Technology
We are an engineering software company based at ULUTEK Technopark, inside the Bursa Uludağ University ecosystem. We turn academic optimization methods into validated products that fit into an engineer's daily workflow.
OptamLab
Decision, design of experiments and optimization platform. Currently in early access.
Research
The methods come from peer-reviewed work, not from marketing.
Assoc. Prof. Emre İsa Albak
Founder · Engineering optimization researcher
Peer-reviewed work on structural lightweighting, crashworthiness, multi-cell thin-walled structures, biomimetic design, thermal management and building design optimization. Neural-network prediction and optimization, surrogate-based multi-objective optimization and multi-criteria decision methods such as grey relational analysis became the recurring tools across that work. OptamLab grew out of a simple observation: the methodological groundwork rebuilt from scratch in every one of those studies deserved to become a permanent tool.
Corporate identity
Nilüfer / Bursa, Türkiye
Common questions
Is the product available right now?
OptamLab is in early access. The engines and all four modules are working; access is granted on request for the time being. Write to us for an enterprise demo.
Where is our data kept?
In the European Union (Frankfurt) region. It is processed under KVKK and GDPR; you can export your analyses and delete them.
How can we confirm the results are correct?
108 worked cases taken from peer-reviewed publications can be inspected without an account, in their engine-re-solved form. Every analysis also carries a configuration hash that guarantees the same input returns the same result.
Does the AI invent results?
No. Every numerical result is computed by deterministic engines. The AI only drafts the prose describing those results, and the numbers in the report are audited against the analysis output.
Can we use our own simulation data?
Yes. You can upload your experimental or simulation results as a table, build a surrogate model on them, and run the optimization over that model.
Which languages are supported?
The interface, reports and support are available in English and Turkish, and reports can be produced in either language.
Do you have a problem worth optimizing?
Describe the problem and let us look together at how OptamLab would approach it. We are here for enterprise demos, academic collaboration and early access.
If the form does not work, write to us directly: info@optam.tech