Build a RAG-powered recommendation chatbot with Qdrant and OpenAI
Give customers a recommendation assistant that understands nuance, including what they want and what to avoid, and returns grounded suggestions from your own catalogue. Built on retrieval-augmented generation, so answers stay accurate and on-brand.
The flow
How the integrations connect.
GitHubHTTP RequestAI AgentOpenAI EmbeddingsOpenAI ChatMemory BufferToken SplitterTool: WorkflowDocument LoaderQdrant
Tools used
10 integrations
Built on n8n. Same pattern works on Make or Zapier for simpler runs, or on a custom Node or Python service when reliability and volume justify the build.
- GitHub
- HTTP Request
- AI Agent
- OpenAI Embeddings
- OpenAI Chat
- Memory Buffer
- Token Splitter
- Tool: Workflow
- Document Loader
- Qdrant
Detail
What it actually does.
- Ingests your product or content catalogue and embeds each item into a Qdrant vector store
- Accepts natural language requests including preferences and exclusions like 'similar to X but not Y'
- Retrieves the most relevant catalogue entries using semantic search over OpenAI embeddings
- Routes results through an AI agent that ranks and explains the top recommendations
- Keeps conversation memory so follow-up questions refine the shortlist rather than restart it
- Grounds every answer in your real catalogue data to prevent fabricated suggestions
- Exposes the recommender as a chat interface ready to embed in your product or site
Common questions
Before you book a call.
Answers to what most teams ask when they look at a workflow like this. If yours is not here, ask us on the call.
Can you build this it ops workflow for our team?
Yes. We design and ship workflows like this as part of our AI Chatbots practice. The fastest way to scope it is a 30-minute call — we share what we would build, what it would cost, and how fast it would ship.
What tools does this workflow use?
The default build connects GitHub, HTTP Request, and AI Agent, plus 7 other integrations. The same pattern works on n8n, Make, or Zapier for simpler runs, or as a custom Node or Python service when reliability and volume justify the build. See Workflow Automation for how we choose the right platform per use case.
What does a build like this typically cost?
Most workflows of this complexity sit inside a Discovery sprint or a small Build engagement rather than a fixed-price product. See our pricing model for how engagements are structured, or book a call and we will scope this specific workflow against your stack.
Have you shipped something like this for clients?
Yes. See our case studies for examples of automation and AI builds we have delivered, including a podcast platform we took from zero to 261K monthly Google impressions in six months on a content + automation engine.
What other IT Ops workflows can you build?
Plenty. Browse our other IT Ops workflow ideas for documented patterns, or tell us what you would like to automate — most clients arrive with a problem rather than a specific workflow in mind.
Want a workflow like this in your stack?
30-minute call. We share what we would build, what it would cost, and how fast it would ship.