We offer consultancy in quantitative finance and AI, alongside bespoke AI
products built to order. We work end to end, from raw data to a system running in
production. Engagements are small, senior and hands-on.
Quantitative research
Systematic strategy development on futures and equities, evaluated to the
standard a trading desk applies.
Feature engineering and signal construction
Walk-forward validation
Execution cost and capacity modelling
Portfolio construction and risk sizing
Applied AI
Machine learning and language-model systems built for real workloads rather
than demonstrations.
Model selection, training and evaluation
Language-model agents, retrieval and tool use
Benchmarking and failure analysis
Deployment, monitoring and cost control
Data infrastructure
The layer everything else stands on: ingestion, storage and pipelines that stay
correct as they grow.
Market and time-series data pipelines
Data quality and outlier detection
High-performance compute in Rust and Python
Reproducible research environments
Approach
How we work, and why.
Data
We keep the raw record rather than a cleaned summary, so that a cleaning
decision can be revisited when a result looks wrong. That is what the
pipelines and quality checks are for. Over 200 TB and counting.
Quant
Anything that might carry predictive power is considered, not only price and
market data. The target is diversified returns rather than one strong signal,
and that governs how signals are combined and sized.
AI
Automation is worth the effort where a task is repetitive and well defined.
Machine learning and language models are applied on that basis.
Quality
Reproducibility and correct data are what a result rests on. Any figure can
be regenerated from the code and data behind it, and traced back to where that
data came from.
Our own AI products.
Ephemera builds and runs its own AI products. Each one came out of our own research
work and is in daily use.
Research automationAgent-driven tooling that runs, verifies and documents quantitative
experiments.
Data quality engineAutomated detection of corrupt prices, corporate actions and pipeline
faults.
Evaluation harnessRepeatable measurement for machine-learning and language-model
systems.
Contact
Tell us what you are trying to find out.
We take on a limited number of engagements at a time. The best first message is a
short description of the problem and what a good answer would let you do.