Neuromorphic Inference Lab Neuromorphic Inference Lab

NLP · Financial Intelligence

GLiNER Financial NER

Zero-shot named entity recognition for banking documents. Detects PII, financial identifiers, and compliance-relevant entities using a DeBERTa encoder fine-tuned on financial corpora and quantised to INT8 for CPU serving.

Source Proxy health
GLiNER ONNX INT8 FastAPI Docker Zero-shot NER PII Detection GDPR / AI Act

Problem → Impact

Financial institutions process millions of unstructured documents daily — contracts, SWIFT messages, KYC forms, transaction memos. Extracting structured entities (IBANs, names, tax IDs, amounts) manually is slow and error-prone. This system provides a production-grade NER microservice with a 24-label taxonomy covering PII and financial identifiers, deployable on commodity CPU hardware.

Live Inference Demo

Entity extraction via a same-origin Cloudflare proxy. The backend runs GLiNER with an ORT INT8-quantised encoder on CPU. Confidence threshold is configurable below.

0.40