Modeling and analysis

Models to decide with data, validated with your team and measured in production

We build quantitative models when the decision depends on computation: simulating scenarios, quantifying a risk, allocating resources, or anticipating demand. We work with your business experts, validate every model against your data, and deliver it ready to run at the latency the case requires, not as a test notebook.

Request the free 30-minute discovery call

Who it is for

  • You have a recurring decision that is made by eye today and enough data to model it: prices, risk, resource allocation, demand.
  • Your system already computes something, such as a simulation, a pricing engine, or a model, and you need it faster, more accurate, or explainable.
  • You need demanding numerical computation with validated results, such as a physical simulation or a language to express correlations.

It is not for you if…

  • you expect a model without data to validate it: first we check whether the data supports the question;
  • you want a test notebook with no path to production;
  • you are looking for a generic artificial intelligence tool: we build the model your decision needs.

Team members’ experience

Work by team members outside SIMYL, before or alongside the company:

  • Oil production simulation (2016–2018): a simulation backend from the wellbore to the separator and a domain-specific language for reservoir correlations (PVT and fluid properties): entry.
  • Monte Carlo engines in trading (2024–2026): optimization of simulation engines to cut execution time; no public entry for confidentiality.

Problems we can evaluate with your specialists

Techniques we master and apply with the people who know the business; they are not a catalog of delivered cases.

  • Monte Carlo simulation and risk analysis: scenarios, outcome distributions, and decisions under uncertainty, including the calculation of probabilities and prices in betting or financial markets.
  • Combinatorial optimization: routing, resource allocation, scheduling, and problems on graphs and flow networks, with integer programming or heuristics when the size demands it.
  • Numerical simulation: physical or process models with calibration and scenarios, and domain-specific languages to express correlations.
  • Forecasting and predictive models: time series and statistical or machine-learning models when the data supports them; we prefer the simplest model that answers the question.

What it takes to validate a model

  • A business question and a success metric agreed before modeling.
  • Enough data and access to the people who know it: we validate with your data and your specialists, not with assumptions of our own.
  • A baseline: how the decision is made today and with what result, to measure the improvement with the method in How we measure.
  • A defined execution environment: latency, volume, and recalculation frequency, because the model is delivered ready for production, with monitoring and recalibration when it applies.

What we deliver

  • An evaluation report with assumptions, validation, and limits, even when the conclusion is that modeling is not worth it.
  • The model in production, documented, with its tests and its monitoring: in C++, Go, or Rust when latency matters; in R or Python for analysis and prototypes.
  • Handover to your team: how it is validated, how it is recalibrated, and when it stops being valid.

How we start

  1. Free 30-minute discovery call: the decision, the available data, and how it is decided today.
  2. Fixed-price evaluation: whether the data supports the question and which model fits, with a defined deliverable.
  3. Build and validation: by milestones, with your specialists and your data.
  4. Delivery: model in production, documentation, and handover.

Terms

Those in How we work: a custom diagnostic for the evaluation and a milestone project for the build, with written acceptance criteria.

Request the discovery call or write to us at contact@simylresearch.com.