Onboarding Programme
These are the typical sessions we take a corporate client on.
Standard Level Support#
Before the first onboarding call#
HumanFirst will setup a dedicated organsation for you and apply the subscription limits you signed up for. You'll get invites to the application and support portals as well as a dedicated slack channel to communicate with the HumanFirst team.
Session 1 - Orientation and project planning#
60 minutes
A general orientation of key labelled and unlabelled data concepts, how to manage your models, subscription, users and organsation. This session uses a test data set and aims to leave you comfortable with how to bring your data into the tool in the best way prior to the second session.
Dataloading#
Prior to second call ensure you have been able to upload your data. The HumanFirst team can provide additional help please send an example of the data in the current format and can work asynchronously to help you via your slack channel.
A guide to the tools available for data loading is available here.
Session 2 - Build and strengthen models#
60 minutes
This session will take you through using a hierarchical classifier to understand your model and data better. How to use clustering, semantic simiarity, search and filtering to find where your model is weak and strengthen it using advanced active learning loops.
Session 3 - Discovering and reducing model confusion#
60 minutes
How to use the NLU evaluation model to accesss the definition of concepts within the model and improve it's quality, accuracy and coverage.
Session 4 - Entities#
60 minutes
Covering how to discover and design high quality entities, how to test and improve entity models and how and when to use them along side intents.
Premium and additional optional sessions#
The four first onboarding sessions cover being profecient with the key areas for model improvement. Optional sessions cover more advanced topics that may be helpful depending on team objective and size. For Premium support users, a custom set of sessions will be developed based on individual need, but this provides and example framework.
Session 5 - Blindset Testing#
30 mins + discussion
What is and how to do blindset testing and maintain your blindests alongside and to the same standard as your train sets in the HumanFirst Studio
Session 6 - Detailed Model Review#
45 mins + discussion
This consists of a detailed review of the current model peformance by the HumanFirst team against the loaded data with recommendations to improve
Session 7 - External Comparisons#
45 mins + discussion
Continuing from the model review session any improvements or updates are reviewed and then further changes are considered based on comparing current peformance with alternate updated technology within the HumanFirst tool. This will be dependent on the target NLU, but might include for instance:
- comparing current RASA pipeline against a set of HumanFirst recommended benchmark pipelines
- comparing Standard, Advanced and Cognitive Playbook options within DialogFlow
- comparing different levels of speech to text quality and mistranscriptions from Whisper, Deepgram and Google off the shelf based on provided input audio
Session 8 - Multi Developer Workflows#
30 mins + discussion
Use of full versioning, merge and tracking of NLU models to integrate with multi environment (QA, Dev, Prod etc) and multideveloper on the same NLU model workflows.