ResearchHub Ethiopia
An AI-powered research discovery platform for aggregating institutional repositories, enabling semantic search and supporting evidence-grounded research assistance.
- Role
Sole author — harvesting, semantic search and the web application
- What it does
Harvests institutional repositories over OAI-PMH and makes them semantically searchable.
- Built with
- Python
- FastAPI
- PostgreSQL
- pgvector
- SQLAlchemy
- Celery
- Next.js
- TypeScript
- Sentence Transformers
- Docker
Overview
ResearchHub Ethiopia aggregates scholarly output from Ethiopian universities into one searchable catalogue. It connects to institutional repositories over OAI-PMH or DSpace REST Discovery, or ingests validated XML, JSON and CSV imports, normalising records while preserving metadata provenance so every entry can be traced back to the repository it came from. Harvesting supports dry runs, full and incremental modes, checkpoints, retries and cancellation. Discovery is semantic as well as lexical: 384-dimensional embeddings are stored in PostgreSQL with pgvector, which also powers related-publication suggestions, and full-text indexing extends search into downloaded PDFs. AI features run against either an OpenAI-compatible provider or a local Ollama model. The repository documents its own maturity honestly — several areas are marked partially implemented, and the README records what has not yet been validated at runtime.
What it does
- Repository connectors for OAI-PMH and DSpace REST Discovery, plus validated XML, JSON and CSV imports
- Harvesting with dry runs, full and incremental modes, checkpoints, retries and cancellation
- Metadata normalisation that preserves provenance for every record
- Semantic discovery over 384-dimensional embeddings stored in pgvector
- Full-text indexing of downloaded PDFs
- Pluggable AI providers, including a local Ollama profile
- Repository health monitoring and import review workflows
Want the detail behind this? Ask me about it.