Where applied ML earns its keep.
MLmargin is our first product, focused on supply chain. The same framework — probabilistic optimization, deep learning where it pays, classical ML where it doesn't — extends into other operational domains. Below are areas we're built for and open to applied-research engagements in.
Detect what signature databases miss.
Cyberattack volume grows 15% per year. Intrusion-detection systems that match against known signatures miss new attacks and drown analysts in false positives. ML closes both gaps.
- Anomaly detection that flags unusual behavior without a pre-existing signature.
- Reduced false alarms — fewer alerts, higher signal per alert.
- Data-loss prevention models trained on positive and counter-examples to classify sensitive content reliably.
Our framework processes thousands of variables per request — fast enough to be in the loop, accurate enough to be trusted.
Forecast load on a destabilized grid.
Renewables make supply less predictable; demand keeps rising. The cost of a wrong forecast is operational, not just commercial — at the limit, grid destabilization. Forecasting models are the right place to spend ML budget in this sector.
- Short-term (minutes) to long-term (weeks) horizons, individual consumer to country-level aggregation.
- Probabilistic outputs — not point estimates — so operators can plan around uncertainty.
- Deep neural networks for the hard cases, classical methods where they suffice. Speed and scalability across both.
Smart grids work when forecasts work. We're interested in being part of that infrastructure.
Compress the discovery loop.
Drug development is slow and expensive — millions of compounds tested, a handful progressing. ML methods are well-suited to the early stages: predicting target structure, biological activity of new ligands, ADMET profiles, and ranking candidates for prioritization.
- Support vector machines, random forests, decision trees, neural networks — picked per problem, not per fashion.
- Anomaly detection, classification, regression — applicable across hit discovery and lead optimization.
- Built to take over the data-science workload from pharma and cosmetics teams that would rather focus on the chemistry.
Shorter cycles between concept, development, and approval mean more treatment options, sooner.
Working on something similar?
Tell us about your domain. If our framework fits, we'll say so. If it doesn't, we'll tell you that too.