The First Neural Generative Learning Platform
Ingest data on Boltzbit platform from different sources and in different formats.
- Supported Platforms: AWS, GCP, Azure or upload
- Supported formats: csv/tsv, json, hdfs(avro, parquet), relational database, nosql database
- Supported modes: batch data, stream data
Model training and evaluation
Orchestrate end-to-end pipelines to build ML solutions.
- full support of the ML development lifecycle
- AutoML without writing and debugging code
- Support Ad hoc and scheduled pipeline executions
- Keep track of pipelines' executions and generated artifacts
Deploy your best model with customized options.
- Online service API type: REST, RCP
- Online service type: CPU, GPU
- Extended service configurations: cloud, resource, traffic load, etc.
- Downloadable model code and artifacts for offline usages.
Register for our beta test now.
Neural network architecture search
Deep Learning is powerful but not well supported on plain AutoML platforms. Boltzbit platform offers cost-efficient Automated Deep Learning.
- Basic AutoML platforms find an optimal model by going through and comparing multiple ML models with different hyper-parameters, which is time consuming and costly. The support of neural networks is limited to exploit powerful deep learning.
- On Boltzbit platform, on top of classic AutoML models, users are offered with cost-efficient Automated Deep Learning via network architecture recommendation and search.
One model. Unlimited Tasks.
One model for one predictive task is expensive and not scalable. Neural Generative Learning builds a versatile model for unlimited tasks.
- Conventional AutoML platforms are task oriented. One model is trained to serve a single predictive task, which is specified by inputs and outputs.
- On Boltzbit platform, users can leverage Neural Generative Learning to build a single model for unlimited predictive or generative tasks.
Built-in Training Data Synthesis
Boltzbit platform integrates DataOps and MLOps. No need to generate synthetic data deliberately.
- Data processing and augmentation are usually not supported at MLOps platforms. Users need to switch between data platforms and MLOps platforms in the development lifecycle.
- Boltzbit Platform is an all-in-one platform to cover both DataOps and MLOps. By virtue of Neural Generative Learning, training data can be easily augmented and synthesised without extra cost and effort.
ML should not be exclusive to Data Scientists. We democratise AI to whoever is passionate about innovation.
- Elementary AutoML platforms are designed for data scientists, which arguably limits the innovative collaboration across teams.
- Besides the coverage of standard data science use cases, Boltzbit platform also provides developer-friendly gears to support software engineers to integrate ML work through their workflow. Of course, we do open source.
Support Cutting-edge AI Research
AI research without the right tool is dull with tedious coding and experimenting. Boltzbit platform simplifies AI research with code generation and experiment management.
- Significant time of ML research is spent on building experiment infrastructure. Researchers are often distracted from working on innovative ideas.
- Boltzibt platform enables researchers to focus on the most critical part by eliminating all coding and experimenting burden. Level up teams' R&D capacity with simple clicks and drags.
Full Deployment Support
Shipping ML outcomes in production is often undervalued on unseasoned AutoML platforms. Boltzbit platform offers wide options for online and offline ML deployment.
- ML model deployment is usually restricted to limited options, e.g. tensorflow, scikit-learn (python). This deficiency hinders ML productionisation when more flexibility is needed.
- Boltzbit platform provides full deployment support in all major programming languages. One can flexibly customize online serving configs or download tailored artifacts for offline usages.