What is a RAG chatbot?
A normal AI chatbot only knows what it learned from the public internet. A RAG (Retrieval-Augmented Generation) chatbot first searches your own knowledge — policies, contracts, manuals, tickets, product data, SQL tables — and then writes an answer based only on what it found. Every answer can link back to the exact page or record it came from, so your team can check it.
That makes RAG the right choice when accuracy matters: legal and compliance teams, healthcare providers, support desks, finance and internal IT helpdesks.
What we build
- Internal knowledge assistants — staff ask questions in plain English and get answers from SharePoint, Confluence, Notion, Google Drive or PDF libraries.
- Customer support chatbots — trained on your help centre and product docs, embedded on your website or WhatsApp, with hand-off to a human.
- Document & contract Q&A — search across thousands of contracts, case files or policies and get cited answers in seconds.
- SQL / database assistants — ask questions about live PostgreSQL or MySQL data without writing queries.
- Multi-tenant RAG for SaaS products — each of your customers only ever sees their own data.
How we build it
- Free audit (week 1). We review your data sources, security rules and the questions users actually ask.
- Data pipeline. Documents are cleaned, chunked and embedded into a vector database (pgvector or Qdrant). Live sources are kept in sync automatically.
- Retrieval tuning. Hybrid search (keyword + semantic), re-ranking and metadata filters so the right passage is found — this is where most cheap chatbots fail.
- Guardrails & permissions. Role-based access control, “I don't know” behaviour when the answer isn't in your data, and citation on every reply.
- Evaluation. We test against a set of real questions and measure answer accuracy before launch, not after.
- Deploy & monitor. Hosted in your AWS account or ours, with logging, cost tracking and monthly improvements.
Private by design
Your documents never need to go to a public AI service. We can run the whole stack — embeddings, vector database and the language model itself — inside your private AWS VPC, or use a private endpoint such as AWS Bedrock with no data retention. See private LLM deployment for fully self-hosted options.
Technology we use
How much does a RAG chatbot cost?
It depends mostly on how many data sources you have, how accurate it must be, and where it's hosted. We've written a full breakdown: How much does a RAG chatbot cost? Every IRT project starts with a free audit and a fixed quote.
Frequently asked questions
How is a RAG chatbot different from ChatGPT?
ChatGPT answers from general internet knowledge. A RAG chatbot searches your own documents and data first, answers only from what it finds, and shows the source for each answer.
Will our data be used to train public AI models?
No. We host the system in a private environment — your AWS VPC or a no-retention private endpoint — so your data is never used to train public models.
How long does it take to build?
A focused knowledge assistant on one or two data sources usually takes 3–6 weeks. Larger, multi-source or self-hosted systems take longer; you get a timeline after the free audit.
Can it stop the chatbot making things up?
We reduce hallucinations with tuned retrieval, re-ranking, strict prompts and citations, and the bot is set to say it doesn't know when the answer isn't in your data. We test accuracy on real questions before launch.
Which file types and systems can it connect to?
PDF, Word, Excel, web pages, Notion, Confluence, Google Drive, SharePoint, PostgreSQL, MySQL, CRMs and most systems with an API.