AI Knowledge Assistant for Your Organisation

We build you a specialised AI chatbot that knows about your data

• Get definitive answers from your data in seconds
• Responses include reference links to the original source documents
• Multilingual by default
• EU hosted and aligned with the EU AI Act
• Data protection compliant with EU regulations such as GDPR

The Process

Data discovery

We start with the subject matter: What kinds of questions do you want to ask? Where would we find the answers? Typically the questions are subject-specific and come from challenges that you are experiencing in your day to day work. The objective is typically to make overcoming these challenges effortless.

We collaborate with you and your team to collect all the relevant data sources. These can be internal sources such as intranet portals, shared drives, or document collections. They might also include public sources on the internet, such as wikipedia, or time-critical sources such as data feeds. Non-text sources, such as video/audio sources may also play a role.

Policies for data access and availability are defined alongside processes to import the data, including preprocessing and indexing. The initial assumptions are defined to specify the importance of different sources.

Prove the concept with a Prototyping or MVP

A fully functional chat webapp is deployed on a customisable link on your own domain for sharing with your initial users. This can be secured behind login or public, depending on your needs.

Here the basic ingredients of a knowledge assistand are in place. Mechanisms to link users back to the data source are developed - citation approach to fit your needs.

We help you to define your strategy to transition to real-world use. This starting point is often to identify a closed group of users with a variety of touchpoints with the subject matter. These users can receive early access to the prototype/MVP. Often the tool will already provide value to them in their work, but it wont be perfect, and that's where their input will prove invaluable to the development process.

Evaluation, Accuracy, Bias

During this initial phase, we will go deeper into concrete examples. Which answers are helpful and which are less helpful? This subject-specific data can be captured and used to optimise the quality of output.

We also use this process to strike a perfect tone when giving answers, including linguistic style, creativity, formatting of replies, and other aspects which affect the usage.

Furthermore, we can use this process to identify and resolve other unwanted behaviours such as bias.

Deploying to production

At this stage refinement of the user interface will also be performed, such as integrating graphs, diagrams, and allowing file uploads.

Data pipelines need to ingest the most accurate and up to date information and must be established for resilience and ongoing updates.

Integrating everything learned so far, we develop a tailored deployment strategy for your production version. Here, decisions such as finalising the Quality Assurance processes, logging preferences, audit trailsm, and how to deploy updates and improvements are all covered. Production will be heavily checked for bugs and monitored to ensure consistent availability.

Get in touch to find out more:

Other Services