Mangaku - Manhwa & Manhua Recommender
An AI manhwa and manhua recommender across 81,000+ titles, fusing semantic embeddings with weighted structured-tag matching in a hybrid Qdrant retrieval pipeline. Every result explains which tags it shares with your query.
The problem
Readers describe what they want in plain language — "regression revenge story with an overpowered lead" — but catalogue search only matches titles and tag filters only match exactly. Pure semantic search solves the language problem and creates a new one: results arrive with no explanation, so there is no reason to trust any of them.
The solution
A hybrid Qdrant retrieval pipeline fuses semantic embeddings with weighted structured-tag matching, so a query is matched on meaning and on the concrete tags it implies rather than on one or the other. Every recommendation is returned with the tags it shares with the query, which makes the result auditable instead of a black box. Three ways to search run on one backend — by title, by description, or by chat. Built solo, end to end, with React/TypeScript as a PWA, FastAPI, Qdrant and OpenAI.
The result
Natural-language search across 81,000+ manhwa and manhua, with every result carrying its own explanation of why it matched. Three search modes share a single backend and retrieval pipeline, so there is one index to maintain rather than three.
Skills used
- AI
- ML
- React
- TypeScript
- FastAPI
- Qdrant
- OpenAI
- PWA
