We build models that learn from noisy signals, reason under uncertainty and extract structure from large, time-varying datasets. Then we build the software that runs them in production.
Figures from our vehicle journey reconstruction programme, development lab, 2026. Read the case study.
Most analytics failures are not modelling failures or engineering failures. They are the gap between the two. We close it by having the people who derive the likelihood also write the pipeline that evaluates it.
Bayesian inference, multivariate and skew-normal likelihoods, covariance estimation, stochastic processes, graph and geospatial methods, information theory. Models built to be validated, not just fitted.
Our methodsData pipelines that handle tens of gigabytes a day, map-matching and routing services, evaluation harnesses, privacy-aware aggregation, and the deployment work that keeps all of it running.
How we build
Hierarchical and sequential models, posterior computation by MCMC and variational methods, principled priors from domain knowledge and historical data.
Skew-normal and heavy-tailed families, covariance and precision estimation in high dimensions, population models that respect the shape of real distributions.
Community structure, flows on networks, association and linkage problems where identifiers rotate and observations fragment.
Map-matching, route likelihoods, spatial point processes, travel-time priors and congestion-aware models of movement.
Entropy and mutual information as design tools: what a signal can and cannot tell you, before you build on it.
Forced-choice evaluation, rank and margin metrics, reliability diagrams, abstention thresholds tuned to the cost of being wrong.
A materials and logistics client needed origin and destination patterns for heavy vehicles across a region. Commercial probe data arrives with rotating identifiers, gaps and five different schemas. We built the statistical linkage and the pipeline to run it at scale.
Read the case study
The methods travel well. The common thread is a large, noisy, time-varying dataset and a decision that depends on getting the uncertainty right.
We take on a small number of engagements at a time. Tell us what you are measuring, what you need to decide, and what the data looks like today. We reply within two working days.
Email
info@advcipher.com
Office
IFZA, Dubai, United Arab Emirates
Working with clients in the UK, Europe and the Gulf.
What to include
The decision you need to support, the data sources you have (or want), timescales, and any constraints on privacy or deployment.