Context and problem

A large media company needed to bring scattered internal knowledge into a single interface. Employees were looking for answers across several sources; searching through the jungle of intranet sources slowed down their day. The goal was an assistant usable in Slack that retrieves information from documents and answers users' questions based on those documents.

What was done

I designed and implemented an AWS-based RAG solution: the database and document index was an AWS Bedrock Knowledge Base, the retrieval logic, language model use, and answering were packaged as Lambda services, and the system was integrated into Slack. The infrastructure was modelled as code (AWS CDK).

I was responsible in particular for retrieval and the AI functionality — scoping the sources, indexing, and prompting, as well as the Slack features (user feedback and analytics). The solution combined several internal information sources into one searchable whole instead of the user wandering around the intranet looking for the information they needed.

Key technologies: Python, AWS Lambda, Bedrock knowledge bases, RAG, LLMs, IaC (CDK).

Outcome

The assistant was deployed in Slack and supported day-to-day information retrieval. The solution showed that an internal multi-source RAG can be deployed on managed infrastructure and integrated into an existing communication channel without a separate portal.

← Back to assignments