*Automated customer services are only as good as the language classifier they use
Our approach to working with clients and partners
Our deep expertise in machine learning and computational linguistics has enabled us to develop the most advanced language classifiers available today – both for speech and text. We tailor these to our client’s specific needs and provide framework source code for your dialogue manager and integration to your application.
With our wealth of experience in design, development, and integration, our team understands the challenges associated with large-scale automation. We provide you with a fully integrated language classifier and a dialogue manager that you can maintain without expertise; your in-house programmer will quickly get up to speed with the dialogue manager code.
Stage 1: Definition and design
Our language classifiers are finely tuned to their targeted domain of use, so, in collaboration with our client, we define the key user functions they wish to automate by voice or text. This task involves:
- Mapping out example user requests or commands and creating model replies. We also help ensure that the relevant parameters are available in the client’s APIs.
- Analysing any available speech or text training data available.
- Discussing options for generating core data or supplementary data in the event of its absence.
- Generating from the data a language ontology for the machine learning classifiers.
- Embarking on the labelling and training process.
Stage 2: Development
Develop: We use our source code frameworks to create dialogue manager source code that suits our client’s language classifier and application. This includes their chosen channel connectors and typically operates on their services, providing them control of their data and service. Tone of voice can be adapted within the dialogue manager at any time by the client, as can a number of other dialogue related changes.
Emulate: The client’s language understanding technology is created in parallel with the dialogue manager, and is integrated on completion of the dialogue manager. In order to create dialogue manager source code before a first version of the classifier is completed, we use a classifier emulator to code and test against. We then connect the actual classifier on completion.
Test: Thorough testing is performed throughout to ensure successful delivery.
Communicate: Continuous communication, and reference to user functions and specifications mitigates mutual risk.
Stage 3: Integration, test and support
We integrate pre-agreed APIs into the dialogue manager and the client can make subsequent changes directly through dialogue manager source code.
Ongoing development resource is available if needed, however our objective is to ensure our client is empowered and self-sufficient.
Proof of concept, pilots and production
Proof of concept applications are often created as the foundation for a customer pilot or full production system.
We generally create a proof of concept for an internal evaluation, building on this for either a live pilot or production system.
