A developer-tools company has 80,000 pages of documentation, API references, changelogs, community forum posts and support tickets. Their current search is keyword-based and users complain constantly: searching "why is my build slow" returns nothing useful, while searching the exact error string works fine.
They want semantic search that handles natural-language questions while remaining excellent at the exact-match cases their users depend on: error codes, function names, config keys.
The corpus is heterogeneous: a 40-line API reference entry and a 6,000-word tutorial are both "documents". Content changes daily, and there are 12 versioned releases of the product with meaningfully different APIs.
Search handles 400 queries per second at peak.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
A user searches for the exact error code 'ERR_CONN_4021'. Why might pure vector search miss the page that contains it?
How do you chunk a 40-line API reference entry and a 6,000-word tutorial differently, and why does it matter?
Do you need a dedicated vector database here? Justify your answer with numbers.
Minimum 6 components · needs a wide desktop screen