Skip to content

AI service

Knowledge assistants that answer from your documents and show their sources

Assistants that answer from your documents and cite their sources.

Start a project

The problem

Where teams get stuck.

Your knowledge sits in contracts, policies, manuals, wikis and old tickets. People cannot find what they need, so they ask a colleague or guess.

A general-purpose model does not know your documents and will fill the gap with something plausible. That is not acceptable for legal, financial or operational questions.

Retrieval-augmented generation fixes this by looking up the relevant passages first and answering only from them. The quality depends almost entirely on how well the retrieval is built.

Our approach

What makes this work.

  1. 01

    Audit the content

    We review what you have, where it lives, how it is structured and who is allowed to see it.

  2. 02

    Build ingestion and retrieval

    Parsing, chunking, embeddings and search are tuned to your documents, including long files, tables and scans.

  3. 03

    Ground the answers

    The assistant answers from retrieved passages, cites them, and declines when the sources do not cover the question.

  4. 04

    Evaluate and launch

    We score retrieval and answer quality on a set of real questions, then launch with monitoring and a way to refresh the index.

Delivery process

How we deliver it, step by step.

Every engagement follows the same seven stages, from requirements to post-launch support.

  1. 01

    Discovery & requirements

    Workshops with your stakeholders to define goals, users, scope and constraints, captured in clear written requirements.

    • Goals and success measures
    • Requirements and user stories
    • Scope, estimate and roadmap
  2. 02

    UX & UI design

    User flows, wireframes and interface design, reviewed with you until the product feels right before it is built.

    • User flows and wireframes
    • Interface design
    • Clickable prototype
  3. 03

    Architecture & planning

    Data model, system design, integrations and security are settled, and the work is broken into sprints.

    • Architecture and data model
    • Technical decisions on record
    • Sprint plan
  4. 04

    Development

    Short iterations with code review on every change and a demo of working software each week.

    • Weekly demo
    • Shared board and written updates
    • Reviewed, version-controlled code
  5. 05

    QA & testing

    Manual and automated testing across functionality, devices, APIs and performance, followed by user acceptance testing.

    • Test plan and test cases
    • Automated regression suite
    • UAT support and sign-off
  6. 06

    Deployment & launch

    Release through a CI/CD pipeline onto monitored cloud infrastructure, with a go-live checklist and rollback plan.

    • CI/CD pipeline
    • Monitoring and alerting
    • Go-live checklist
  7. 07

    Hypercare & support

    Close monitoring and priority fixes straight after launch, then ongoing support, maintenance and improvement.

    • Hypercare period
    • Support under agreed response times
    • Improvement backlog

Deliverables

What you receive.

  • Ingestion pipeline for documents, PDFs, wikis and tickets
  • Vector and keyword search tuned to your content
  • An assistant that cites the passages behind each answer
  • Permission-aware retrieval, so users see only what they may access
  • An evaluation set for retrieval and answer accuracy
  • Re-indexing so new and changed documents are picked up
  • Usage analytics and a log of unanswered questions

Technology

What we build it with.

  • OpenAI
  • Anthropic
  • Google Gemini
  • Python
  • FastAPI
  • PostgreSQL
  • pgvector
  • Supabase
  • Next.js

Related work

Where we have done this before.

From the blog

Related guides.

What we have written about rag & knowledge assistants: cost, timelines and the trade-offs behind each decision.

FAQ

RAG & Knowledge Assistants: common questions.

What is RAG?

Retrieval-augmented generation. Relevant passages are fetched from your own content and passed to the model, so the answer is based on your documents and can cite them.

Do you train a model on our documents?

No. The content is indexed and looked up at question time. There is no training run, and updating the assistant means re-indexing content, not retraining.

Can it handle very long documents?

Yes. Long contracts and reports are split into structured sections and retrieved in pieces. We have built a multi-agent pipeline for long legal contracts on this pattern.

Can it get things wrong?

Yes. No system removes errors completely. Grounding, citations and refusal behaviour reduce them, and the evaluation set tells you how often they happen. Citations let users check any answer.

Can we control who sees which documents?

Yes. Retrieval respects your existing permissions, so an answer is only built from content that user is allowed to read.

Discuss your RAG & Knowledge Assistants project with us.

Share a short brief and we will reply with next steps, usually within one working day.

Start a project