Case 01 — Nexus ERP & OPUS Multi-Tenant SaaS
Autonomous ingestion and OCR extraction engine processing heterogeneous municipal gazettes and healthcare bids with deterministic schema validation and multi-tenant isolation.
Deconstructs PDF pages, feeds vision tokens into Gemini 1.5 with prompt schemas for high-density tabular price extractions.
Processing heterogeneous, multi-page public procurement gazettes previously demanded 45+ minutes of manual human inspection per document. The volume of bids across municipal portals created extreme backlogs, high error rates in pharmaceutical price mapping, and prohibitive cloud costs if raw documents were fed naively into commercial multimodal models.
Engineered a shared PostgreSQL cluster with strict Row-Level Security (RLS) instead of isolated databases per tenant. This eliminated baseline AWS RDS overhead, simplified atomic schema migrations across 100+ entities, and consolidated backups. Paired with asynchronous Celery/Redis workers to ensure Gemini LLM extraction latency (4-8s) never blocks the synchronous React/TypeScript user interface.