Machine Learning and Deep Learning
Budgets and forecasts that come from your history, not from intuition.
Sound familiar?
- Budgets are done by hand and depend on who does them.
- Cost or schedule deviations show up when they can no longer be fixed.
- You have years of history that you do not use to make decisions.
What you get out of it
- You will know the cost before you start
- You will plan with data, not intuition
- You will catch deviations in time
- You will decide on measurable criteria
Predictions built on your real data.
We train machine learning and deep learning models on your company's historical data and leave them working inside your tools, with a confidence margin on every prediction.
Cost forecasting
We estimate the cost of work items or projects from your real history, with a confidence interval on every result.
Demand and lead-time forecasting
We anticipate workloads, consumption and delivery dates so you plan with data, not intuition.
Anomaly detection
We flag deviations in invoices, consumption or budgets before they cost money.
Ranking and classification
We score opportunities, risks or suppliers with measurable criteria you can review.
How it is integrated and validated
It is delivered as a web tool or an API connected to your ERP or spreadsheets. It is validated by measuring its accuracy on data it has not seen and comparing it with how you estimate today.
We clean and join your historical data.
We test several models and keep the one that works best.
We measure its accuracy on data it has not seen.
We put it into service and monitor it.
How we do it
What you get
- 01Trained and documented model
- 02Ready-to-use tool, with a confidence margin
- 03Accuracy report and when not to trust the model
- 04Retraining plan
Where it runs
We decide with you where the system lives, depending on how sensitive your data is.
On your servers
We install it on your infrastructure or in your cloud. Your data never leaves your environment; in exchange, someone on your side has to maintain those servers.
On Vidra's servers
We host it on servers inside the European Union and you access it via web or API. We take care of maintenance, and access is restricted to your team.
With external AI APIs
For language models we use the OpenAI, Anthropic, Google or OpenRouter APIs. It is the most capable option and the fastest to set up. The text fragments needed are sent to that provider, and we tell you which one is used and what data it receives.
With local models
If your data cannot leave your environment, we use models that run entirely on your servers or ours, without sending anything to third parties. They are somewhat less capable than the large API models and need more hardware.
From pilot to production
We start with a scoped pilot: one specific case, with your real data and measurable goals agreed up front. If they are met, we integrate it into your tools and it goes into production. If not, we explain why and what it would take.
After delivery
When the system goes into production, a feedback stage begins: your team uses it day to day and, if something fails or does not fit, we fix it. After that you can hire maintenance: incident handling, updates and, for models, retraining with new data.
What you hire and what we build it with
What you hire
A forecasting system running in your tools. It starts with a pilot and accuracy goals agreed before work begins.
Under the hood
A predictive model trained on your history, used as a web tool or an API.
- scikit-learn
- XGBoost
- CatBoost
- LightGBM
- PyTorch
- FastAPI
We choose the tools for your case; you do not need to know them.
