HumanFirst is not an NLU engine, but rather a centralized data layer on top of existing conversational AI platforms that makes it easier to discover, build, train, maintain and refactor intents and their training data.
HumanFirst is platform agnostic in regards to the NLU engine you are using.
Our system tolerates typos and mistakes because it uses sub-word tokens that will be resilient to typos and grammar errors.
How do we get metrics / F-score / perplexity metrics ?
Our pipeline does a stratified k-fold split and extracts key metrics for workspaces (f1, precision, recall)
We don’t expose those just yet, but we do have them on the short term roadmap to help guide the development of intents.
On top of that, we do plan to expose even more active learning features to streamline intents training process even more.
Can we manage multiple bots/agents structure?
We have the concept of a workspace in which you can develop your intents structure. You can create multiple workspaces in a single account, which can be used to manage different projects within the same organization.
Conversational data can be shared across multiple workspaces.
For companies managing projects for various clients, HumanFirst allows for distinct namespaces unique to each organization/client.
Data uploaded to HumanFirst is stored under Google Cloud's infrastructure. Our data pipeline processes the uploaded file and creates derived representation that can power features like semantic search and recommendations. Data is always encrypted at rest using the AES-256 cipher, and industry-standard security best practices are followed. For more information about Google Cloud's security practices, visit https://cloud.google.com/security
Who can see the data I upload?
When you sign up, you are asked to create an organization, you can then invite teammates to this organization in order to share workspaces and unlabelled data sets. People you invite can immediately interact with the data in your organization. You can also revoke access at any time by deleting the user from the settings section.
If I delete an uploaded file, is it really deleted?
Utterances and conversation files are immediately deleted following the action from the user interface. Some components of our infrastructure maintain cached copies of the data in order to provide fast access to recommendations and search results - it may therefore take a few minutes before deleted data disappears from a workspace. Automated backups to prevent accidental data loss are kept for an additional 30 days and are automatically cleaned up without needing manual intervention.
Are you using my data in other ways?
We will from time to time run algorithm benchmarks and monitor the quality of recommendations provided in order to improve the service. In no case whatsoever do we make your data available to any third parties.
How is PII handled?
We expect PII information to be redacted from the uploaded contents, if you notice that you have accidentally uploaded PII data, simply delete the affected file. If you legally require some confirmation of deletion, we will gladly provide one upon request.