JNS-SRC-01
AI-Powered Search Engine
Semantic search that understands the question, not just the keywords.
1 Overview
A search engine that embeds both the query and the corpus, blends semantic and keyword scores, and returns a short synthesized answer above the classic result list.
2 Block diagram
- 01EmbedQuery to vector
- 02SearchVector + BM25, fused scores
- 03SummarizeAnswer from the top results
3 Pin-out
- QUERYtextSearch phrase or question
- RESULTSranked listDocuments with highlighted passages
- SUMMARYtextShort synthesized answer with links
4 Specifications
| Parameter | Value | Notes |
|---|---|---|
| Latency p50 | 320 ms | results; answer streams after |
| Index size | ~120k chunks | — |
| Cost per query | €0.0006 | — |
| Recall@10 | 0.91 | internal benchmark |
5 Stack
- Models
- Embedding modelSmall fast LLM for summaries
- Data
- Vector indexInverted keyword index
6 Limits
- Summary is skipped when top results disagree strongly.