Case studyConfidential client projectAI/ML2025Architecture & build

: RAG-powered AI customer support

Production-grade RAG-powered AI support platform with vector search, multi-LLM support, and intelligent content recommendations for Athlean-X customer support.

I need something like thisAsk for a walkthrough
LLM providers
3
Hybrid rank fusion
RRF
Vector stores
2
Confidential client projectAI/ML · 2025
Private client build
AX AI Support Platform

Screenshots and source are withheld at the client’s request. I can walk you through the architecture and the decisions behind it on a call.

3
LLM providers
RRF
Hybrid rank fusion
2
Vector stores
01 — The project

The challenge

Support teams answer the same questions every day, and the answers are spread across articles, videos, and product pages. A chatbot that guesses is worse than none: customers need answers grounded in the company’s own content, with links to the right resources. The platform also had to work with more than one AI provider, so cost, speed, and quality could be balanced per task without rewriting the system each time a better model appears.

Designed and implemented a sophisticated AI-powered support system using RAG (Retrieval-Augmented Generation) architecture. Built a hybrid search pipeline combining Qdrant vector database with PostgreSQL pgvector for semantic search, implemented Reciprocal Rank Fusion (RRF) for improved relevance, and created a multi-LLM streaming interface supporting Ollama, OpenAI, and Claude models.

Highlights

  1. 01RAG pipeline with Qdrant + pgvector for high-accuracy semantic search.
  2. 02Hybrid search with Reciprocal Rank Fusion (RRF) for improved relevance.
  3. 03Multi-LLM streaming support including Ollama, OpenAI, and Claude.
  4. 04Admin dashboard with analytics, prompt versioning, and content management.

What this means for your business

If your business has years of help articles, FAQs, or documentation, the same approach can turn them into an assistant that answers customers accurately and points them to the right page. What matters is the same at any size: search that finds the right source, answers grounded in it, and an admin view that shows what people ask. I can scope a smaller version for your site, from a smarter search box to a full support assistant.

Stack

  • Next.js
  • FastAPI
  • Python
  • Qdrant
  • PostgreSQL
  • Redis
  • Docker
  • LLMs
  • RAG
  • OpenAI
  • Claude
  • pgvector
  • Vector Search
  • Embeddings
  • Prompt Engineering
Back to all work
02 — Keep reading

More work

Need something like this?

  • Web applicationsCustom software when a website is not the answer: multi-tenant SaaS, billing, dashboards and internal tools.

Related guides

03 — Contact

Tell me what you’re working on and roughly when you need it. You’ll hear back within one business day.

contact@cevallossystems.comReplies within one business day

Replies within one business day