AI for Processing Large Documentations

How we made a dedicated server with AI for easy search through documentation and developed search algorithms.

Client

WorkFusion is the leading provider of AI digital workforce solutions for Fortune 500 enterprises, banks, insurance, and financial services companies.
Our company has been working with WF for many years on various projects and tasks. This project is a training school for working with these products.

Сhallenge

The main task was to help the user quickly find the necessary information during training. Each product has more than 10 versions and the documentation for each of them is very large. Products closely intersect with each other, and it is quite challenging to understand structural connections in documentation.

Solution

Our solution is a dedicated server with AI connected to the basic search engine. We have also developed a new search result page that makes it easier to work with each individual user’s queries. At the system level, counters of various kinds of requests and lists of the most popular questions from users are implemented.

AI Search & Results

We’ve seamlessly integrated an AI assistant into Moodle’s search, providing users with comprehensive support. This AI is adept at analyzing documentation and delivering precise answers to platform users. Not only does it address inquiries directly, but it also furnishes relevant links to specific sections within the documentation. This enhancement speeds up navigation through various documents by product and version, promoting a more user-friendly experience.

Key Clever Integration Features

Generating Correct URL Links

Navigating through extensive documentation can be challenging, especially when faced with markdown and HTML formats. Recognizing this hurdle, we developed a solution to streamline comprehension. By systematically removing HTML and markdown tags, we distilled the content into its pure textual form. This approach empowers AI to generate accurate links, ensuring users are directed to relevant resources without confusion.

Training Generative Model

Crafting an effective AI requires a meticulous training process. We conducted a thorough evaluation, comparing the performance of Google’s AI cores with ChatGPT. Each core underwent intensive training on our dataset, enabling them to discern patterns and relationships within the data. Through this comparative analysis, we identified the core that exhibited superior performance, tailored to our requirements.

Automation of Model Retraining

Continuous improvement is imperative in the realm of AI. To facilitate this, we implemented automated retraining processes. Our AI autonomously initiates retraining based on predefined criteria, such as the availability of new data or performance metrics. This automated approach eliminates the need for manual intervention, ensuring our model evolves seamlessly over time.

Statistics Collection Functionality

Understanding user behavior is essential for refining AI system. Through admin panel’s statistical collection feature, we gain valuable insights into user interactions. Armed with this data, we can optimize documentation development, enhance search functionality, and improve overall user experience. By leveraging these insights, we strive to meet the evolving needs of our users effectively.

Feedback and Communication

Comprehensive communication strategy has boosted our project efficiency. We continually meet client needs and expectations by keeping everyone informed, aligned and empowered to deliver results.

Microsoft Teams & Jira

We prioritize clear and efficient communication. In the process of work, we always discussed updates, as well as approved all stages of work with the client through calls to Microsoft Teams. This fostered open communication and allowed for immediate feedback. We utilized Jira to track tasks, issues, and project progress. Team members could provide detailed feedback within Jira tickets, ensuring clarity and accountability.

Results

The client received unique AI in its Moodle platform, which will help users obtain information easily and quickly. Our solution will also help the client continue to improve both the products themselves and their documentation.

Deliverables

A fully functioning separate server with a trained request processing model, that will automatically retrain based on changes in the documentation.
Integration into the search bar on the site and a dedicated page for the user to work with AI and its control.
Pages for managing preset questions.
Pages for collecting statistics on user questions.

Results in figures

People

330

Hours

15 K+

Pages of documentation
processed

Technologies and tools

Python, Flask, Google Cloud Platform, Chroma DB, Google Generative AI.

WorkFusion & SmartApp Partnership

For over seven years, our company has enjoyed a successful and fruitful partnership with WorkFusion, a leading provider of AI-powered digital workforce solutions. Our long-term partnership with WorkFusion is a testament to our shared values and dedication to innovation. We look forward to continuing this successful collaboration and contributing to WorkFusion’s continued growth and success.

Ready to start?

Contact our CTO or fill out the form

Sergei Porfenovich