Skip to content
dhii.ai

Assistants, Chat & Voice

RAG & Knowledge Assistants

An assistant that answers from your own documents and data — citing its sources, respecting who is allowed to see what, and saying "I don't know" when it should.

Pilot on one source
2–4 weeks
Multi-source production
6–10 weeks
Typical team
1–2 engineers

The knowledge problem

Every organisation has the answers somewhere — in a policy PDF, a Confluence page, a closed ticket, a spreadsheet on someone’s drive. Finding them is the problem. New staff ask colleagues; colleagues get interrupted; customers wait. Traditional enterprise search returns ten links when people wanted one sentence.

A well-built RAG assistant returns the sentence — and the link to where it came from.

Retrieval is where projects succeed or fail

Most disappointing RAG systems have the same root cause: the right passage was never retrieved, so the model answered from nothing. We treat retrieval as its own engineering problem and measure it separately:

  • Parsing that respects structure. Tables stay tables, headings stay attached to their sections, scanned pages get OCR.
  • Chunking per document type. Policies, product manuals and support tickets need different strategies.
  • Hybrid search. Vector similarity finds meaning; keyword search finds exact product codes and names. We use both, then re-rank.
  • Metadata filters. Region, product, date and permissions narrow the search before the model sees anything.

Only once the right evidence is reliably found do we tune how answers are written.

Grounded, cited, honest

Answers include citations to the exact passages used. When evidence is missing or contradictory, the assistant says so and points to the closest sources or a human contact. Conversation memory is handled carefully so follow-up questions work without leaking context between users.

Measured before and after launch

We build an evaluation set from real questions — not invented ones — and score retrieval recall, answer correctness and faithfulness. After launch, thumbs-up/down feedback and unanswered questions feed a weekly review that usually reveals content gaps as much as technical ones. Fixing a missing policy page often helps more than changing a model.

Deliverables

What you get

  • 01

    Knowledge source audit

    Inventory of documents, wikis, tickets and databases with quality, freshness and permission notes.

  • 02

    Ingestion pipeline

    Connectors, parsing, chunking and embedding with scheduled re-sync and deletion handling.

  • 03

    Retrieval layer

    Hybrid keyword and vector search with re-ranking, metadata filters and per-user access control.

  • 04

    Answering service

    Grounded answers with inline citations, refusal when evidence is missing and conversation memory.

  • 05

    Evaluation set

    Hundreds of real questions with expected answers, scored for retrieval and answer quality.

  • 06

    Interfaces

    Web widget, Slack or Teams bot, or an API for your own app.

Process

How a rag & knowledge assistants project runs

  1. Step 1

    Collect questions

    Gather the real questions people ask, from tickets, chat logs and interviews.

  2. Step 2

    Ingest

    Connect and clean the sources, and design chunking for your document types.

  3. Step 3

    Tune retrieval

    Measure whether the right passages are found before working on answer wording.

  4. Step 4

    Ground answers

    Add citations, refusal behaviour and guardrails, then evaluate end-to-end.

  5. Step 5

    Launch & learn

    Release to a pilot group, capture feedback and fix gaps in content and retrieval.

Engagement models

Ways to work together

Every engagement starts with a fixed, written quote. See engagement models & pricing or the how much does a rag chatbot or knowledge assistant cost?.

FAQ

RAG & Knowledge Assistants: common questions

What is RAG and why not just fine-tune a model on our documents?

Retrieval-augmented generation looks up relevant passages from your content at question time and asks the model to answer from them. It keeps answers current, lets you cite sources and respects permissions. Fine-tuning changes a model's behaviour and style but is a poor way to teach it facts that change. Most knowledge assistants should use RAG; some benefit from both.

How do you stop the assistant making things up?

By retrieving the right evidence, instructing the model to answer only from it, showing citations, and refusing when evidence is weak. We measure faithfulness on an evaluation set and track it over time. No system is perfect, so we design the interface to make sources easy to check.

Can it respect document permissions?

Yes. We carry access-control metadata from the source systems into the index and filter retrieval by the user's identity, so people only get answers from content they could already open.

Can it run entirely on our own servers?

Yes. Open-weight embedding models, a self-hosted vector store and a local LLM via vLLM or Ollama give a fully private deployment.

Which document types do you support?

PDF, Word, PowerPoint, Excel, HTML, Markdown, Confluence, Notion, SharePoint, Google Drive, ticket systems and SQL databases, plus scanned documents with OCR.

Related services

Start a project

Have a workflow that should run itself?

Tell us what you want automated or built. You get a written assessment of feasibility, risks and a fixed quote — usually within two business days.