Our mission is to understand each and every action as accurately as possible from all available data. To predict behaviors. To make the right choices. To help you care.

Join innovative organizations in using BaseModel.ai (Synerise)
as a core AI framework to query the future with multiple questions
in such segments like: retail 🔗 telco 🔗 finance 🔗 and more.
Asics - BasemodelIKEA - BasemodelOrange - BasemodelCompany logoCompany logoCompany logoCompany logoCompany logoZabka - BasemodelCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoCompany logoNike - BasemodelNike - BasemodelNike - Basemodel
General
How do daily customer interactions influence their future behaviors?
Customer service
What is the customer’s likelihood of using a special offer?
Insurance
How many insurance policies will the customer subscribe to this year?
Gaming
How many power-ups/bundles will the gamer buy this month?
Banking
What is the customer’s projected profitability in the next quarter?
Travel
What is the customer’s expected number of trips this year?
Health
How many diagnostic tests will the patient need this year?
Telco
How much data traffic will the customer use this month?
E-Commerce
Which products/promotions/ offers/categories the customer is interested in?
Automotive
What kind of product/category is the customer interested in and why?
Home & Furniture
How to split the customer population into behaviorally distinctive groups?
Payments
What is the utility of customer for your business and what are the factors affecting it?
Fashion
Will the customer make a purchase next week?
Fashion
What steps need to be taken to increase the chance of purchase?
Security
Is recent behavior of the customer inconsistent with past habits?
Compliance
Are there outlier customers in the population, who might be worth looking into?
News & Publishing
Will the reader subscribe to a premium plan?
Why?

Predicting behaviors is a key component of the future

Rapid shifts caused by the exponential growth of AI capabilities will cause profound ripples in the fabric of the economy, society and civilization. Humanity forms a global community, yet every person is unique. Societies function cohesively because we can build mental models of other people.

Mental models allow us to predict what makes other people happy, what behaviors are best avoided, what is amusing, and what is boring. What motivates, and inspires, and causes boredom, discomfort, or pain? Knowing how someone else might react is the foundation of our civilization. As humans, we do that naturally, but our abilities did not evolve to scale beyond a few dozen people within our closest communities. Caring about millions or billions of people is a super-human task, and we use AI to solve it.

By analyzing the full spectrum of observable information about people - their historical interactions with content, actions, decisions, and choices, we build foundation AI models to predict individual future behaviors in real and hypothetical scenarios.
Jack Dąbrowski
Chief AI Officer Synerise / Founder Basemodel.ai
Organization

Understanding people and predicting their behaviors at scale is hard

Many companies have teams of Data Scientists & ML engineers but the complexity and availability of data grow exponentially. They struggle with:
Lack of time and capacity to understand and utilize all available data
Evolving data schemas and new incoming data sources
Creating, maintaining and debugging data pipelines
Handcrafting features describing behaviors
Very high dimensional observation & decision spaces
Each model requires different features and/or input formats
Guaranteeing fundamental correctness (no data leakage, proper temporal splits)
People

Human ability to manually wrangle data scales sub-linearly.

All of this leads to:
A hard cap on the number and complexity of predictive models
A lot of spaghetti code, impossible to grow, hard to maintain
Entrenchment in a "data science" silo, with arcane knowledge and little results to show
Employees becoming complacent and defending their "arcane ways"
State of the now

The ways how Data Scientists, Data & ML/AI Engineers work with behavioral data today are severe bottlenecks.

Understanding complex & intricate patterns of interactions is a super-human challenge. Imagine a single model could learn from all your raw data. Such a model could form a foundation for solving any applied task with unparalleled efficiency and quality.

Manual labor & lack of time

Each ML project required careful manual labor, starting with analyzing available data sources. Lack of time and capacity to understand and utilize all available data

Expert knowledge

Evolving data schemas and new incoming data sources. Countless handcrafted features had to be created using expert knowledge

Human limitations

Despite best efforts, important behavioral cues were often lost due to human limitations. Handcrafting features describing behaviors.

Complicated data pipelines

Creating, maintaining and debugging data pipelines is hard taks.

Multiple models

Each model requires different features and/or input formats.

Data bias

The information content of raw data was orders of magnitude richer than the actual input of models. Very high dimensional observation & decision spaces.
What

BaseModel.ai reduces modeling life-cycle to days instead of months & eliminates common data engineering problems and supercharges behavioral ML.

A single data scientist becomes supercharged 10x 🚀

Automatic feature creation & foundation model training from raw data

No more messy pipelines
No more unused / underutilized data sources
No more problems with evolving data sources
No more maintenance work for keeping the foundations in shape

BaseModel learns to extract, link and interpret information from multiple data sources

No longer constrained by simplistic human heuristics
No more challenges with "multi-modal" or "multi-dataset" fusion
No more months-long trial & error on "what features to use for model X"
No more need for manual feature creation

Fine-tuned models trained just by defining the objective

No more custom projects for each use-case
No more plumbing for every single model
All models benefit from all the data
No more mistakes and problems with information leakage
No more arcane ways of training for specific purposes

Deepest explainability on the market

Identify singular high impact events
Pinpoint exact behavioral patterns driving predictions
Assign scores to every single interaction in raw data

Current standard approach to ML

The current outdated standard approach to machine learning often involves manual feature engineering and model selection, which can be time-consuming and less accurate. It also typically relies on batch learning, which is incapable of adapting to new data in real time, limiting its flexibility and responsiveness.

BaseModel.ai approach to ML

Modern approaches to machine learning leverage automation, with techniques such as deep learning and autoML reducing the need for manual feature engineering and model selection.
Woman looking at phone

Dailymotion.com has applied EMDE/BaseModel.ai to personalize video recommendations in native applications, leading to improved relevance and catalog coverage

EMDE (BaseModel.ai) gives us a generalised framework for recommendations. The embedding generation was superfast (i.e <5 minutes). For context, do remember that GraphSAGE took ~20hours for the same data in the NCR region.

BaseModel.ai is core for all AI services offered in Synerise platform

BaseModel’s powerful behavioral analysis enables us to make sure that our clients receive communication tailored specifically to their preference.

Just 5 days: from configuration to first trained model

Basemodel.ai encapsulates the rapid and streamlined process of setting up and training a machine learning model with modern tools and techniques. This expedited timeline, which includes data preparation, model configuration, training, and evaluation, highlights the efficiency and agility of current ML methodologies.
Features

No more „plumbing pipelines” and „handcrafting model inputs”. Ideate, prototype, evaluate!

Thanks to BaseModal.ai, applied models train faster, require less labelled data and perform better than classic approaches.
BM models are trained 2x faster than baselines
BM needs 3.5x less labelled data than baselines
BM’s quality metrics significantly outperform classic baselines
BM performs advanced feature engineering, representation learning & training automatically
Dashboard mockup
Benefits

Your private foundation model can bootstrap various applications

Just like large language models can be fine-tuned to any application, BaseModel.ai is not restricted to a predefined set of use cases – you can unleash your creativity freely.

Propensity prediction

Uncover the hidden desires of your customers. BaseModel can accurately predict human inclination towards services, products, brands, categories, styles, or other fine-grained attributes, helping you strategize your marketing approach and increase sales.

Churn prediction

Stay ahead of human attrition with our churn prediction capabilities. BaseModel estimates the likelihood of a customer discontinuing your services or products at any level of granularity, enabling you to take preventative actions and maintain your client base.

Recommendations

Craft personalized human journeys. BaseModel offers individualized recommendations on services, products, brands, styles and any fine-grained attributes you may find relevant - enhancing human satisfaction and loyalty.

Anomaly detection

Spot the unexpected early! On an individual level, BaseModel identifies unusual changes in user behavior, helping you respond promptly to potential issues or opportunities. On an aggregate level, users with atypical behavior patterns can be easily spotted for further analysis.

Profile scoring

Understand your customers better than ever before. BaseModel estimates customer loyalty, lifetime value, their propensity to engage with your communications, or any other metric you come up with - ensuring you can tailor your approach to maximize business success.

Profile matching

Find your perfect match! BaseModel helps identify the most suitable customer groups for campaigns, upsells, and cross-sells, ensuring your efforts reach the most responsive audience.

Hyper-segmentation

Segment like never before. BaseModel can divide your customer base into distinct groups based on behavior, enabling hyper-targeted marketing and improved customer relationships. Forget about manually defining segments based on socio-demographic attributes - let behaviors speak for themselves!

Time to event predictions

Stay one step ahead with BaseModel's time-to-event predictions. Whether it's anticipating the next purchase or predicting when a customer will top up their account, our AI keeps you in the know.

Uplift estimation

Find your perfect match! BaseModel helps identify the most suitable customer groups for campaigns, upsells, and cross-sells, ensuring your efforts reach the most responsive audience.

Unprecedented accuracy of behavioral modeling to ML

BaseModal.ai looks at all human interactions with your organization and learns to predict future behaviors. Automatically. At scale.
Integrations

Simple, secure and quick deployment, with zero overhead

We’ve made sure that starting the journey with BaseModel.ai is smooth,
can be deployed on your own cloud subscription, or on-premise (NVIDIA GPU-equipped server required)
available as a hardened Docker image. It doesn’t require clusters, multiple micro-services or standalone databases
does not duplicate your data. It reads the necessary data directly from your data-warehouse in a streaming way, during model training & inference
easily auditable and can pass the most stringent security requirements.
Integration iconIntegration iconIntegration iconIntegration iconIntegration iconIntegration icon
Case study

BaseModel.ai is blazingly fast.
No clusters needed

Our mission to equip every organization with a private behavioral foundation model makes us very conscious about resources. That’s why we take cost-efficiency seriously & optimize our algorithms to be lightning-fast and extremely scalable.
A full self-supervised training pipeline for 10M+ customer profiles with 1 year worth of data can complete within a few hours on a VM equipped with a single GPU. Subsequent fine-tuning for supervised applications is even faster, and requires very little labeled data.
Research

Challenging top AI powerhouses

We take science seriously. We verified BaseModel.ai power at the most prestigious research venues, where early versions of BaseModel.ai took the podium competing with global giants like Google DeepMind, Baidu Research, NVIDIA, Intel, Oppo Research, Xerox PARC, Rakuten and 800+ teams from leading universities worldwide. The difference? Our competitors spent weeks hand-crafting their solutions for every problem separately, with zero reusability. We just applied BaseModel.ai to all the problems.

Solving applied problems becomes a piece of cake.

Solving applied problems becomes a piece of cake thanks to understanding the full spectrum of behaviors. The foundation model is trained only once, and then adapted to specific tasks in a process called fine-tuning. Not having to relearn behaviors from scratch, a fine-tuned model only needs a target to zoom in on.
Careers

What we offer:

very high impact - changing how organizations work, globally
defining the cutting edge of AI and ML
a very different perspective, very different kind of data & problems from all the other AI labs
extremely bright and experienced team
very high organizational flexibility
competitive compensation
100% remote work, possibility to work in a hybrid mode, or on-site in any of our office locations (Poland: Warsaw, Krakow, US: San Francisco)
the scientifically strict, but caring and kind atmosphere

What we expect:

you did not stop learning after graduating
you can both generate ideas and execute them to completion
you are not ashamed of communicating your needs, problems and concerns
you judge ideas based on their merit as opposed to their source, authority or way of communication
you are not afraid to fail, but instead treat failures as "training data" for self-improvement

We are sharing our ideas with others!

Our research papers based on BaseModel.ai framework
Multidimensional Hopfield Network
Redefining Graph Clustering: A Convergence of Algorithms and Networks
Temporal graph models fail to capture global temporal dynamics
We propose a trivial optimization-free baseline of "recently popular nodes" outperforming other methods on all medium and large-size datasets in the Temporal Graph Benchmark.
A Foundation Model for Behavioral Event Data
SIGIR '23: Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2023
Real-Time Multimodal Behavioral Modeling
CIKM '22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management, 2022
An Efficient Manifold Density Estimator for All Recommendation Systems
International Conference on Neural Information Processing (ICONIP 2021)
Cleora: a Simple, Strong and Scalable Graph Embedding Scheme
International Conference on Neural Information Processing (ICONIP 2021)
Twitter User Engagement Prediction with a Fast Neural Model
15th ACM Conference on Recommender Systems RecSys Challenge Workshop, 2021
Node Classification in Massive Heterogeneous Graphs
ACM's Special Interest Group on Knowledge Discovery and Data Mining (SIGKDD) KDD Cup Open Graph Benchmark (OGB) Challenge Workshop, 2021
Efficient Manifold Density Estimator for Cross-Modal Retrieval
The 43th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) eCom Workshop, 2020
Modeling Multi-Destination Trips with Sketch-Based Model
14th ACM International Web Search and Data Mining Conference (WSDM) WebTour Workshop on Web Tourism, 2021
On the Unreasonable Effectiveness of Centroids in Image Retrieval
International Conference on Neural Information Processing (ICONIP 2021)
Interpretable Efficient Multimodal Recommender
Thirty-seventh International Conference on Machine Learning (ICML) Machine Learning for Media Discovery (ML4MD) Workshop, 2020

We are sharing our ideas with others!

Our research papers based on Synerise BaseModel.ai framework
Multidimensional Hopfield Network
Redefining Graph Clustering: A Convergence of Algorithms and Networks
A Foundation Model for Behavioral Event Data
SIGIR '23: Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2023
Real-Time Multimodal Behavioral Modeling
CIKM '22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management, 2022
An Efficient Manifold Density Estimator for All Recommendation Systems
International Conference on Neural Information Processing (ICONIP 2021)
Cleora: a Simple, Strong and Scalable Graph Embedding Scheme
International Conference on Neural Information Processing (ICONIP 2021)
Twitter User Engagement Prediction with a Fast Neural Model
15th ACM Conference on Recommender Systems RecSys Challenge Workshop, 2021
Node Classification in Massive Heterogeneous Graphs
ACM's Special Interest Group on Knowledge Discovery and Data Mining (SIGKDD) KDD Cup Open Graph Benchmark (OGB) Challenge Workshop, 2021
Efficient Manifold Density Estimator for Cross-Modal Retrieval
The 43th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) eCom Workshop, 2020
Modeling Multi-Destination Trips with Sketch-Based Model
14th ACM International Web Search and Data Mining Conference (WSDM) WebTour Workshop on Web Tourism, 2021
On the Unreasonable Effectiveness of Centroids in Image Retrieval
International Conference on Neural Information Processing (ICONIP 2021)
Interpretable Efficient Multimodal Recommender
Thirty-seventh International Conference on Machine Learning (ICML) Machine Learning for Media Discovery (ML4MD) Workshop, 2020
Temporal graph models fail to capture global temporal dynamics
We propose a trivial optimization-free baseline of "recently popular nodes" outperforming other methods on all medium and large-size datasets in the Temporal Graph Benchmark.
Science

Lab

Sair is a lab focused on behavioral modeling, recommendations, large-scale data and graphs processing. We share our ideas, models, and experimental results, also presenting our take on important breakthroughs and interesting technologies. We hope to build a better and more thorough understanding of the field. We believe in the importance of this research not only from a business perspective but most importantly as a study of human decision-making processes.
Research
8 min read

BaseModel vs TIGER for sequential recommendations

The comparison between BaseModel and TIGER reveals substantial differences in their architectural choices and performance.
Read post
Research
8 min read

BaseModel vs HSTU for sequential recommendations

To evaluate BaseModel against HSTU, we replicated the exact data preparation, training, validation, and testing protocols described in the HSTU paper.
Read post
Research
8 min read

Fourier Feature Encoding of numerical features

Pre-processing raw input data is a very important part of any machine learning pipeline, often crucial for end model performance
Read post
Future
6 min read

Why We Need Inhuman Artificial Intelligence

We continuously wonder how much longer it will take until AI reaches human skill level in these tasks - or, when does AI become "truly" intelligent.
Read post
Engineering
12 min read

EMDE vs Multiresolution Hash Encoding

When we created our EMDE algorithm we primarily had in mind the domain of behavioral profiling.
Read post
Tools
8 min read

Efficient integer pair hashing

Mental models are simple expressions of complex processes or relationships.
Read post
Research
9 min read

Cleora: how we handle billion-scale graph data

We have recently open sourced Cleora — an ultra fast vertex embedding tool for graphs & hypergraphs.
Read post
Research
8 min read

Towards a multi-purpose behavioral model

In various subfields of AI research, there is a tendency to create models which can serve many different tasks with minimal fine-tuning effort.
Read post
Research
10 min read

EMDE Illustrated

In this article we provide some intuitive explanations of our objectives and theoretical background of the Efficient Manifold Density Estimator (EMDE)
Read post
Research
7 min read

How we challenge the Transformer

Having achieved remarkable successes in natural language and image processing, Transformers have finally found their way into the area of recommendation.
Read post