Rejuve Bio

Rejuve Biotech's longevity research depends on querying complex biomedical relationships (genes, proteins, pathways) stored across multiple knowledge-graph backends - but researchers can't write graph queries, and the upload path for new datasets could only handle 2-3 concurrent uploads.

Build the two neural-symbolic layers around Rejuve's biomedical knowledge graphs: a natural-language annotation query service, and a custom AtomSpace builder that turns CSV datasets into MeTTa/Neo4j/MORK graphs at scale.

Built the multi-backend annotation query service (a Flask API supporting Neo4j Cypher queries, MeTTa symbolic reasoning, and MORK CLI execution) with hash-based query deduplication and a result formatter that normalizes outputs across all backends. Separately, built the custom AtomSpace builder (FastAPI) moving CSV preprocessing to Celery workers with Redis as broker, a shared upload state store, and per-session cache locks - the request persists, enqueues, and returns immediately while workers parse and build DataSource metadata with atomic cache writes.