AI/ML research
We find out whether a machine learning approach will work on your problem before you pay to build it.
- Pillar
- AI
- Typical timeline
- 2 to 8 weeks
- Price
- Quoted in a written scope after the first call
Who it is for
Companies with data and an open question: can a model predict this, classify that, or read these documents well enough to be useful?
The problem
AI projects most often fail at the start, when nobody checked whether the data can support the goal. Months go into building around a model that was never going to be accurate enough.
What you get
- 01A review of your data: volume, quality, labels, and gaps
- 02A baseline: the simplest method that works, to measure everything else against
- 03Experiments with suitable models, from classical machine learning to language models
- 04A written report with results, error analysis, and a clear go or no-go
- 05Notebooks and code so your team can reproduce every number
How it works
-
Step 1
Frame the question
We turn the goal into a measurable target and agree what good enough means.
-
Step 2
Review the data
We look closely at the data before any modelling.
-
Step 3
Experiment
Baseline first, then stronger models, with every run logged.
-
Step 4
Report
Findings, limits, and a recommendation you can act on.
Timeline
2 to 8 weeks
A typical range for this kind of work. The written scope we send after the first call sets your actual dates.
Related work
Projects for this service are being written up. See all work
Questions buyers ask
What if the answer is no?
Then you learn it in weeks instead of months, and the report says what data or change would make it work.
Does our data have to leave our systems?
No. We can work inside your cloud account or on a machine you control.
Will you publish the research?
Only with your permission. By default the work and the results are yours.
Tell us what you are trying to build
A 30-minute call. You explain the problem, we ask questions, and you leave knowing whether we are the right team and what the work would involve.