Sheet 03 — Case Study
AI search infrastructure
LiveRetrievalKit
Self-hostable hybrid search for teams building AI agents and RAG pipelines, running on the Elasticsearch cluster they already operate.
The problem
The usual answer to retrieval for AI is a separate vector database: another system to run, another place for data to drift, and another bill, even for teams that already run Elasticsearch.
Who it is for: Engineering teams that already run Elasticsearch and need retrieval for AI features.
What we built
- Hybrid search that merges keyword and semantic results in one query against one index
- Embeddings computed by Elasticsearch's own inference API, with no external embedding service
- Automatic fallback to keyword search when a cluster has no inference endpoint
- Filters, facets, highlighting and related documents
- A signed-in admin panel that re-indexes a document on every save, and a live demo search
Engineering quality
- A content security policy that allows no third-party host, with fonts and styles served from the site itself
- Automated axe accessibility tests and end-to-end tests
- Analytics load only after a visitor agrees
Tech stack
- Python
- Flask
- Elasticsearch
- Jinja2
- Vercel
Every project is scoped to your goals.
Book a call for a tailored estimate.