Company knowledge that answers back,without making things up.
Our retrieval-augmented business assistant: an in-house R&D build that turns an organisation's own documents into a grounded, conversational aide.
The situation
Every organisation sits on a pile of knowledge — policies, product specs, case files, customer profiles — that only answers questions if somebody goes digging. Generic chatbots answer fast but invent facts; search finds files, not answers.
We wanted a business aide that anyone can talk to in plain language, that answers from the company's own verified knowledge, and that can point to its sources. So we built one ourselves, as internal R&D, before putting the pattern in front of clients.
What we built
RAGbot: a retrieval-augmented generation assistant with three core elements — a database of raw company information, a knowledge base that structures it, and a large language model that does the talking.
After evaluating the AWS and Azure stacks, we built the production path on AWS: Amazon Bedrock as the RAG foundation, S3 for document storage, OpenSearch for vectorisation, Claude by Anthropic as the model, and Lambda functions for workflows and API integration.
Full-stack serverless and deployable on websites and web apps, customised per business: legal teams pulling case files, sales answering product queries on the spot, HR fielding leave and policy questions.
What changed
Answers are grounded in a sourced, verifiable knowledge base, which keeps hallucinations out of the loop and lets people trust what comes back.
Knowledge that used to live in heads and inboxes now flows through a tool anyone can use, automating knowledge sharing and delivering measurable time savings on admin and information hunting.
RAGbot is where we proved the retrieval pattern that now underpins the AI work across our client platforms.
“Think ChatGPT, but with intimate knowledge of your organisation.”