Customer Support

Support Knowledge Base RAG Assistant

Answer support questions from your own documentation, with citations

Featured Verified Advanced Agentic Workflow Mostly automated Paid
Use this workflow
Est. time
~15 min
Est. cost
$0.35
Steps
7
Automation
Mostly automated
Trigger
Chat

Overview

A support assistant that only answers from your documentation. Questions are embedded, the relevant passages retrieved and re-ranked, and the answer composed strictly from what was retrieved — with citations the user can click.

The important behaviour is the refusal: when the corpus does not contain the answer, it says so and escalates rather than inventing one.

Business value

Deflects routine questions with answers staff can audit, instead of a chatbot that guesses.

Expected outcome

Cited answers drawn only from your corpus, with unanswerable questions escalated rather than guessed.

Problems it solves

  • Documentation exists but nobody can find the right page
  • Generic chatbots answer from training data, not your docs
  • No audit trail for what the assistant told a customer
  • Documentation drift goes unnoticed until someone complains

Who it is for

  • Support teams with substantial documentation
  • Internal helpdesks
  • Regulated organisations that need auditable answers

Benefits

  • Every answer cites the document it came from
  • Refuses rather than inventing when the corpus is silent
  • Re-indexes as documentation changes — no retraining
  • Retrieval respects the asking user's permissions

Limitations

  • Answer quality is capped by retrieval quality — bad chunks, bad answers
  • Chunking strategy needs tuning per corpus
  • Multi-hop questions spanning several documents remain hard
  • Keeping the index in sync with the source of truth is ongoing work

Success metrics

  • Deflection rate
  • Groundedness (answers fully supported by citations)
  • Escalation rate and its accuracy
  • Documentation gaps surfaced per week

Implementation guide

Index a narrow, high-quality subset first — your twenty most-viewed help articles. Measure groundedness on fifty real questions before widening the corpus. Most RAG deployments fail because they indexed everything on day one and never measured retrieval.

Flow diagram

Every step in order, colour-coded by how much of it runs without a human.

Scroll to pan · use the controls to zoom, expand or export.

Steps

7 steps from trigger to result.

1

Ingest and chunk the corpus

4 min Automated
Split documents into passages that each stand alone, preserving headings and document identity.
InputHelp centre export
OutputChunked passages
  • Chunk on semantic boundaries, not fixed character counts — a passage cut mid-procedure retrieves badly
2

Embed and index

Pinecone 3 min Automated
Generate embeddings and write them to the vector store with permission metadata attached.
InputChunked passages
OutputVector index
3

Retrieve for the question

1 min Automated
Embed the incoming question and fetch the nearest passages, filtered by what this user is allowed to see.
InputUser question
OutputCandidate passages
  • Filter by permission at retrieval time — never retrieve everything and rely on the model to decline
4

Re-rank

1 min Automated
Re-score candidates for actual relevance to the question, not just embedding proximity.
InputCandidates
OutputTop passages
  • Re-ranking is the single highest-leverage upgrade to a mediocre RAG setup
5

Compose a grounded answer

Claude Sonnet 2 min Automated
Answer strictly from the retrieved passages, citing each claim — or state that the answer is not in the documentation.
InputQuestion + top passages
OutputCited answer
Prompt
Answer the question using ONLY the passages provided.

Question: {{question}}

Passages:
{{passages}}

Rules:
- Every factual claim must cite the passage it came from, as [1], [2]…
- If the passages do not contain the answer, reply exactly: NOT_IN_DOCS
- Do not use knowledge from outside these passages, even if you are confident
- Be direct. No preamble.
  • The NOT_IN_DOCS path must be wired to escalation — a question that silently dead-ends is worse than no assistant
6

Escalate or resolve

2 min Semi-automated
Cited answers go to the user; NOT_IN_DOCS routes to a human and logs a documentation gap.
InputAnswer or refusal
OutputResolution or escalation
Expected result: Every question either answered with citations or escalated with context.
7

Review gaps weekly

2 min Manual Human review
The logged NOT_IN_DOCS questions become the documentation backlog.
InputGap log
OutputDocumentation tasks
Expected result: Documentation improves from real questions rather than guesses.

AI tools

Each tool records the role it plays and whether it is required.

Integrations

Services this workflow connects to, and how each authenticates.

Prompt library

Every prompt in the workflow, in chain order. Swap the highlighted placeholders for your own values.

5. Compose a grounded answer · User prompt · Claude Sonnet
Answer the question using ONLY the passages provided.

Question: {{question}}

Passages:
{{passages}}

Rules:
- Every factual claim must cite the passage it came from, as [1], [2]…
- If the passages do not contain the answer, reply exactly: NOT_IN_DOCS
- Do not use knowledge from outside these passages, even if you are confident
- Be direct. No preamble.
question passages

Prerequisites

Have these ready before you start.

LLM API key
For embeddings and generation.
Vector database
Pinecone, Weaviate, Qdrant or pgvector.
Document corpus
Help centre, runbooks or policy documents in a machine-readable form.
Access-control mapping
Which documents each user role may retrieve from.

Use cases

Customer-facing docs search

Let users ask questions of the documentation and get cited answers.

Reduced routine ticket volume

Internal helpdesk

Answer staff questions from policies and runbooks with links to the source.

Faster internal answers

Regulated advice lookup

Search policy sets and quote the governing clause rather than paraphrasing.

Auditable answers

Metrics

Author estimates — treat them as a starting point and measure your own runs.

Execution time
~15 min
Estimated AI cost
$0.35
Token usage
18K
Success rate
86%
Automated
82%
Human review
18%

Accepts

TextDocumentsPDFAPI

Produces

SummaryEmailDashboard

FAQ

Does RAG eliminate hallucination?
It reduces it substantially but does not eliminate it — the model can still misread a retrieved passage. Clickable citations are what let a user verify, which is why they are not optional here.
Was this helpful?
Why is re-ranking a separate step?
Nearest-neighbour search returns passages that are semantically close, which is not the same as passages that answer the question. Re-ranking is usually the biggest single quality jump available.
Was this helpful?
How do we keep the index current?
Re-index on document change rather than on a schedule. A weekly cron leaves up to a week of wrong answers after a policy update.
Was this helpful?

Reviews

No reviews yet — be the first to share how this worked for you.

Write a review

You will be asked to sign in.