Engineering case study
Production-Grade AI Search Engine
A Perplexity-style research assistant combining fresh web retrieval, source ranking and streaming AI synthesis with inline citations.
01 / EVIDENCE
The problem
Conventional search returns links that users must verify and synthesize. Fluent model answers can be fast but difficult to trust when sources and freshness are hidden.
02 / EVIDENCE
The system approach
A LangGraph pipeline retrieves fresh web material, extracts and ranks sources, then streams a synthesized answer while keeping evidence visible.
- Search → fetch → extract → rank → synthesize
- Inline citations and source panel
- Server-sent event streaming
- Redis caching and rate limiting
- Conversation and collection persistence
03 / EVIDENCE
Key engineering decisions
The design balances retrieval depth against latency, source relevance against domain credibility, and transparency against interface overload.
04 / EVIDENCE
What this demonstrates
End-to-end retrieval product engineering. Grounding and citations reduce unsupported answers; they do not eliminate model error.
Next step
Discuss a similar challenge.
Start with the business context, desired outcome and current constraints. We will establish whether Norrelium is a sensible fit.