ML and Data
I provide outcome oriented solutions within the data space. Whether you need a data product that will handle a huge scale data or a deep dive into modeling the dataset you’ve gathered, I’m here to help.
No dataset? I’ve designed collection protocols guided by the project targets.
Do you have a business goal with just an inkling of how it could be achieved? I offer a thorough analysis of your targets through a practitioner’s lens and will design a system, or depending on the problem, a research plan that leads to such a system.
I’m happy to take on a research problem and have experience improving state-of-the-art published performance within my work, public and private.
On the other end of the application spectrum, I’ve also taken off-the-shelf solutions, adapted them to the task at hand, and created a working system at tight startup timelines for my clients.
My approach is heavily driven by what the intended applications are. As much as I love to create sophisticated ML pipelines, before I embark on creating that involved computer vision system, we’ll first consider using a scanner to make the task easier and feasible with a free OCR package.
A machine learning system is not just the glamorous model architecture. To achieve good real world performance, one needs to answer questions such as:
- Where does the data come from? How often is it updated?
- How often do we need to retrain the adaptive parts? How costly and complex is that?
- Do we need adaptation on faster timeframes than the retrains?
- How often do we need to update the pipeline outputs? What latency is acceptable on it?
- How and where do we output?
- How do we know the system is performing well in production?
During design and implementation I take such issues into account to provide a data system that you can host and run seamlessly, not just Jupyter notebook.
While my deep specialty is data and ML, I’ve created and productionized numerous software systems in my life.
I aim for simplicity, time-to-market and realistic evaluation of needs and means. I’m ruthlessly pragmatic and refrain from holy wars – what works works, and in particular, what matters is what works for you.
I offer fractional CTO services, mostly targeted at startups. Some things I’ve done for my clients within that domain include:
- Designing systems architecture with a strong focus on agility and planned features, as well as the required performance envelope.
- Identifying and fixing bottlenecks in existing systems, performance or cost.
- Introducing modern software development practices (such as code reviews, CI, and a goal dependent level of planning).
- … and as a strong generalist I can fix your backend/database/etc. when it’s called for.
I understand that your technology needs depend on the business you’re running. In some cases the value directly depends on the predictive accuracy and speed of your code, such as in various algorithmic risk taking endeavors. In others, an attractive feature might be what you need to raise additional funds, sell to new customers or upsell existing ones. There are also cases where the lacking responsiveness of your systems might be threatening churn.
All of those scenarios should be treated differently, designed with the purpose in mind and executed at the right point on time, quality and maintainability Pareto front.