March 4, 2026
AI Data Flywheel Demo
Achieving Large‑Model Accuracy with Faster, Cost‑Efficient Small Models.
This demo showcases a real‑world proof of concept that applies the NVIDIA AI Data Flywheel Blueprint to medical coding, demonstrating how smaller, fine‑tuned models can achieve high accuracy at a fraction of the cost and latency of large foundation models.
The solution ingests physician clinical notes and predicts ICD‑10 diagnosis codes, comparing performance across a large 70B reference model, a fast but less accurate 1B base model, and a 1B model refined through the Data Flywheel. By continuously tuning the small model with curated data and evaluation loops, the solution closely matches the accuracy of the large model while delivering sub‑second responses, significantly lower infrastructure costs, and edge‑ready deployment.
Built by WWT in the AI Proving Ground on the Cisco Secure AI Factory with NVIDIA infrastructure, this POC illustrates how enterprises can scale high‑value AI use cases efficiently without massive GPUs, centralized data centers, or excessive operational expense.
Hardware and software:
- Cisco Nexus 400GB Switching
- 1 x Cisco UCS C845A Server
- 4 x Nvidia H200 NVL GPUs
- NVIDIA AI Enterprise
- RedHat Openshift Container Platform
- Meta Llama-3.2-1b-instruct NIM
- Meta Llama-3.3-70b-instruct NIM
- NVIDIA NeMo Data Designer
- NVIDIA Data Flywheel Blueprint