A sea of possibilities. One decision.
predicated.ai turns what might happen into something a person can act on. Forecasting and decision analysis, machine learning, language-model practice, and the engineering chops to match.
Scroll — the field resolves ↓What we get hired to do.
Numbers with the uncertainty still attached
Bayesian models, time series and simulation — and the visualisation that makes the result legible to someone who did not build it. You get a range and a probability with the assumptions written down, so anyone relying on it can see how much weight it will take.
- Bayesian inference
- Hierarchical models
- Time series
- Monte Carlo
- Calibration & backtesting
- Information visualisation
Models trained on your problem
Acquisition, augmentation and pipelines, then training and tuning against your data rather than a public benchmark. Most of the work sits either side of the model.
- Data acquisition
- Ingestion pipelines
- Augmentation
- Bespoke training
- Evaluation & tuning
LLMs, past the demo
Where they earn their keep in engineering, operations and market research: what to hand over, how to check it, and how the answers stay reproducible when the model changes underneath you.
- Agentic engineering
- Evals & guardrails
- Ops automation
- Research at scale
- Team enablement
Backends built for the load they’ll get
Event-driven systems at scale, search and information retrieval, and the data and ML infrastructure underneath. Designed around the traffic and the failure modes that turn up in year two.
- Event-driven systems
- Search & retrieval
- Streaming pipelines
- ML platform
- Scale & reliability
Find the signal. Build the system.
Spread out over the possibilities, then narrow — and arrive with the range, the reasoning and the failure modes still attached. That shape holds whether what we hand over is a forecast, a trained model, an agent that does real work, or a system that has to stay up.
Don’t take our word for it.
Israeli Knesset Election 2026
A public forecast of the 26th Knesset, rebuilt every day: every poll published, house effects measured against past elections, surplus agreements, and twenty thousand simulated chambers. In Hebrew and English, with the method written out and the back-test scores printed next to it.
sklearn-hierarchical-classification
Labels arranged as a tree rather than a flat list — the shape a taxonomy actually has, and the one scikit-learn leaves you to build for yourself. We wrote the estimator that fills the gap and put it out under Apache 2.0.
It did the hierarchical work in the entry that took first place in subtask B of GermEval 2019, over 343 nodes of German book blurbs — chosen ahead of the team’s own implementation, the paper says, so the result could be reproduced.
The library → The paper →PromptRomp
Ten open-source projects, most of them the tool we wanted while building something
else. pytest-impacted runs only the tests a change can actually reach,
by way of git, the AST and a dependency graph. mockstack stands in for
the services a microservice depends on before they exist.
aws-bootstrap-g4dn takes one command from nothing to a GPU box on
AWS, with the PyTorch build that matches the CUDA it found — which is the part
that usually costs the afternoon. pdfalive gives a PDF the table of
contents it should have shipped with, and works out for itself when it has to OCR
the thing first.
Enterprise LLM Design Patterns
A short series on what breaks when language models meet real systems. The first is about answers that arrive stale — a model call is slow relative to the state it is reasoning about — and pairs debouncing with optimistic locking.
The second is routing: deciding which model may see which data when compliance has an opinion about it.
Read the series →