Agglestone Labs

Agglestone Labs · Data science research

Data science research for questions that don't have an answer yet.

Some problems can't be solved by picking a tool off the shelf. You don't know whether the data holds the answer, which method will work, or how well it can work. Finding out is research, and that's the work we do. It suits any field with hard data problems, from science and medicine to finance and logistics.

What we research

Software engineering and data science problems where the method, or whether it can be done at all, isn't known at the start.

Finding the signal in messy data Working out what your data can and can't tell you, whether it's patchy records, free text, sensor readings or years of transactions.
Forecasting and prediction Models that estimate what happens next, and how sure they are. See predictive models.
Text and documents Reading, sorting and pulling facts out of documents, emails and forms with language models, measured against people doing the same job.
Optimisation and decisions Scheduling, routing, pricing and allocation problems where the best answer has to be searched for, not looked up.
New algorithms and systems Making software faster, more accurate or able to cope with data at a scale where the usual approach falls over.
Experiments with real users Testing changes to a product or service with the people who use it, designed and measured so the result can be trusted.

How a research project runs

A systematic approach, written down as we go, so every result can be checked and repeated.

  1. 1Pin down the questionWhat we're trying to find out, and what's uncertain about it: why the answer isn't already known or obvious.
  2. 2Check what's already knownA review of existing methods, papers and tools, so the research goes into the part that's genuinely new.
  3. 3Plan the experimentsA hypothesis for each one, the data it needs, a baseline to beat and how success is measured, agreed before we start.
  4. 4Run, measure, repeatEach experiment is run and compared with the baseline, and what we learn shapes the next one.
  5. 5Record everythingEvery experiment, setting and result goes in the project log, including the ones that didn't work. They're findings too.
  6. 6Hand over what we foundA write-up of the results and the reasons behind them, plus the code and models, ready to build on.

What you get

  • A research plan. The question, what's uncertain, the hypotheses and how each one will be tested.
  • An experiment log. What was tried, when, with what data and settings, and what happened.
  • A findings report. What we learned, how confident we are, and what we'd try next.
  • Working code and models. Everything needed to repeat the results and build on them.

We use only the data a project needs, keep it secure while we work with it, and agree with you up front how it's handled when the project ends.

Got a question your data might answer?

Tell us about the problem and what you've tried so far. We'll tell you honestly whether it needs research, and how we'd go about it.

Talk to us