Return
ITTI - Success Story
Generative AI conversational assistant for secure access to corporate knowledge
A generative AI-based conversational assistant was implemented to facilitate secure access to internal knowledge, improving operational efficiency.
Collaboration
About the Client
ITI is a technology company that develops digital solutions aimed at improving the management and access to information within organizations.
Through its platforms, it allows data centralization and internal process optimization, driving the digital transformation of its clients.
Needs
1
The organization had multiple sources of information, documentation, and services distributed across different systems.
2
This created difficulties in quickly accessing internal knowledge, responding to inquiries consistently, and scaling the use of artificial intelligence without compromising information security.
3
Furthermore, they sought to implement generative AI capabilities without needing to train models with their own data or expose sensitive information.
What was done?
Generative AI Architecture on AWS
A generative AI architecture was designed and implemented on AWS that allows for the deployment of a secure, scalable conversational assistant integrated with the client's systems.
RAG implementation
The solution uses a Retrieval Augmented Generation (RAG) approach, allowing the model to respond solely with information from verified sources.
Deployment was standardized using infrastructure as code:
Security measures, content filtering, and data isolation were implemented to ensure the confidentiality of the information.
Infrastructure as Code and Integration
Deployment was standardized using Infrastructure as Code, integrating multiple services and queries within a single conversational channel.
Results
Efficient access to knowledge
The implementation made it possible to roll out a conversational assistant capable of answering queries with reliable, verified information, significantly improving access to internal knowledge.
Improved response quality
A knowledge base was developed with quality control mechanisms that enable the accuracy of responses to be measured, potential model hallucinations to be detected, and the documentary sources used to be validated.
Continuous and reliable operation
The project was carried out by a cross-functional team using short cycles and continuous reviews, ensuring that the technical expertise remained within the client’s organization.
Streamlining internal processes
Based on the first use case implemented, a replicable template and an automated deployment process were developed, enabling the solution to be scaled to different groups with minimal adjustments and accelerating the time-to-market for new AI use cases.
We went from building models in months to building them in weeks, which allowed us to significantly reduce our time to market. In the processing phase, we managed to reduce our 45% by 45% and then improved our error rate by a factor of 10 thanks to autoscaling and native observability.
Customer Testimonial:
""In my experience working with Nublit, their greatest contribution was guiding us through this process of industrializing the practical application of generative AI. The result of everything we’ve built is a dynamic platform that allows us to scale our products and business models.""
