On June 9, 2026, NVIDIA shared the winning teams and projects from its "NVIDIA Spark Hack Series" held in Toronto during Toronto Tech Week. Participating teams used NVIDIA Nemotron, an open-source model that can run locally on DGX Spark, to build agentic applications that operate autonomously.
Toronto Tech Week · NVIDIA Spark Hack Series
Local-First AI Agents, Built On a Desktop Supercomputer
Teams shipped fully on-device agentic apps on the NVIDIA DGX Spark — pairing open Nemotron models with Toronto's civic data, no cloud required. Four track winners turned private, low-latency AI into real urban use cases.
1 PFLOP
FP4 AI performance (~1,000 TOPS) from the GB10 Grace Blackwell Superchip
128GB
LPDDR5x unified memory shared by CPU + GPU at 273 GB/s
2.6×
Speedup NVIDIA cites for NemoClaw, unboxing-to-agent in minutes
NemoClaw acceleration
Same memory, same chip — building local agents goes from baseline to up to 2.6× faster.
Track Winners — agentic apps running fully on-device
Economic Systems
CityFlow
Fuses traffic, weather, event & 311 data — forecasts demand with Nemotron, optimizes staffing with cuOpt.
Public Services
Belong
On-device companion for dementia patients — conversation, memory, schedule & face recognition stay local.
Urban Operations
FlowTO
A digital twin of Toronto's roads & transit — proposes interventions from plain-language requests.
Best Use of Nemotron
Cracked City
Analyzes uploaded photos of road damage and automates 311 reporting end-to-end.
What developers praised
Unified memory and high bandwidth make running large models like Nemotron locally smooth — practical agents built in a short time, unbox-to-agent in minutes.
The caveats
The ARM64 ecosystem and software compatibility are still early, and a ~$2,999–$3,999 price points it toward professionals and researchers more than hobbyists.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…
