// NVIDIA Inception Capital Connect Member · July 2026
Built in the field.
Proven on the platform.
Sounder Solutions didn't start with a pitch deck. It started with a corral trap, a mesh network, and a landowner who built his own solution a quarter mile from his house in Northeast Texas. Three hardware generations and an NVIDIA Inception acceptance later — Sentinel V3 runs 67 TOPS of edge AI on a DIN-rail mounted Jetson Orin Nano Super, off-grid, no cell signal required.
The technology came before the business. That's why it works.
Act 1 · Before
The trap came before the company.
Before there was a company name, a patent, or a pitch deck — there was a working system. A corral trap, a quarter mile from the house on property in Northeast Texas. Erick ran a mesh network to it himself so he could watch hogs on a screen from his living room and pull the trigger remotely. No AI. No investors. Just a solution that worked because he needed it to.
Manual · Mesh Network
The trap came before the company. A corral, a mesh network run from the house, a screen, and a manual trigger. Quarter mile from the front door. No AI. No investors. Fully operational.
Raspberry Pi 5 + Cellular
V2 proved cellular reach and field autonomy. IR counting worked. But the Coral edge TPU was a ceiling — limited model support, no path to the inference speeds the platform needed.
Jetson Orin Nano Super Current
NVIDIA Inception accepted. Training moved to AWS GPU. YOLOv8 → TensorRT. The V3 runs dual-stream inference at 60–100+ FPS at 1080p. It's not a trap controller. It's a ruggedized field AI.
Act 2 · Now
How we actually use NVIDIA's platform.
Every layer. End to end.
This isn't a logo wall. Every NVIDIA tool listed here is running in our pipeline today or is the named next milestone. Here's what each one does for us — specifically — and why it matters that they work together.
Local Development + Iteration
Where: Development workstation · NVIDIA GeForce RTX 5070 Ti
What: Dataset preparation, label validation, rapid training experiments,
and model debugging happen locally. Small batch training runs, hyperparameter sweeps,
and TensorRT export testing — all without burning cloud credits. When a model
architecture works locally, it goes to AWS for full-scale training.
Why NVIDIA: Same CUDA toolkit, same cuDNN, same TensorRT pipeline
as the cloud and the Jetson. A model that trains on the 5070 Ti exports to the
A10G without translation. One SDK across every device in the stack.
Cloud-Scale Training
Where: AWS G5 instance · NVIDIA A10G Tensor Core GPU
What: Full YOLOv8 training runs on the complete East Texas dataset.
Overhead thermal and visual imagery — not stock datasets. The exact camera angles,
terrain, IR signatures, and lighting conditions that match V3 deployment.
$10,000 in GPU credits through NVIDIA Inception.
Why NVIDIA: The A10G is a data-center class Tensor Core GPU.
Training that takes hours locally runs in minutes at scale. Same CUDA code,
larger batch sizes, faster convergence.
Optimize for the Edge
Where: Post-training (local or cloud)
What: Export the trained model through TensorRT. Quantize to FP16/INT8.
The model goes from a 200MB research artifact to a tight, optimized engine
that runs at 60–100+ FPS on the Jetson. From "hog detected" to "gate fires"
in under one second.
Why NVIDIA: TensorRT is the only optimization path that targets
Jetson hardware natively. It's not a generic quantization tool — it knows
the Orin's architecture and extracts every TOPS available.
Deploy to the Field
Where: Jetson Orin Nano Super · DIN-rail mounted · off-grid
What: The optimized TensorRT engine runs locally on JetPack SDK.
Dual camera streams. Thermal interrupt wake. When the model says "full sounder
inside the corral" — the gate fires. No server call. No cellular dependency.
The decision happens on-device at 67 TOPS.
Why NVIDIA: JetPack SDK is the full-stack Jetson OS — Linux,
CUDA runtime, cuDNN, TensorRT runtime, multimedia API. One platform from
training to deployment. No framework translation. No inference server.
Current status · Honest
What's done. What's next.
We don't claim production-ready at scale. We claim three generations of working hardware, a patent on the trigger logic, AWS training infrastructure, and a clear path to field deployment.
V1–V3 Hardware Architecture Complete
Three generations. V3 hardware architecture finalized. DIN-rail mounted Jetson Orin Nano Super with dual REOLINK cameras, AMG8833 thermal, 150W solar, 100Ah LiFePO4, 4G/GNSS.
Patent Filed · No. 63/783,959
FAF™ (Full-Array Fire) trigger logic. Filed April 5, 2025. Assigned to the LLC. Addresses the primary failure mode of conventional trapping: partial capture.
NVIDIA Inception Accepted · ICC Approved
NVIDIA Inception member. Inception Capital Connect approved July 29, 2026. Active 90-day window — hundreds of VC firms in NVIDIA's ecosystem have access to the Sounder Solutions profile.
AWS Training Infrastructure Live
G5 instance with NVIDIA A10G configured. Training pipeline operational. Dataset rebuild strategy identified — East Texas field imagery prioritized over generic European wild boar datasets.
YOLOv8 Model Training
Initial training run completed on harvested frames. Label quality issues identified (non-target species mislabeled). Dataset cleanup and East Texas field image labeling in progress. Target: 300 labeled overhead thermal images.
Synthetic Data · NVIDIA Omniverse Replicator
TensorRT Export + Jetson Field Deployment
Optimize trained model for Jetson Orin. Target: 30+ FPS per stream, FP16 precision, <50ms inference latency. Validate in real field conditions. Document performance.
Funded Pilot · 5 Production Units
First 5 units deployed on partner ranches. NRCS EQIP cost-share integration. Capture dataset begins building. Pilot LOI from AgriLife or county extension is the highest-value pre-raise asset.
Act 3 · Where We're Heading
Not just better detection.
A proprietary data asset.
The V3 platform wasn't designed to be a smarter trap. It was designed to be a persistent identity system. During NVIDIA Inception onboarding, a strategic reframe locked in: the Jetson Orin doesn't just run object detection — it can run Vision-Language Models locally. That changes everything.
Discrete software layers. Works. Proven on V2/V3 prototypes.
One model. One pipeline. Runs locally on Jetson. No cloud latency.
The moat
Persistent sounder identity. A dataset no competitor can buy.
A fine-tuned VLM enables persistent animal identity tracking — recognize individual animals across sessions, verify whole-sounder capture before the gate fires, build population behavior profiles from real East Texas deployments. NRCS, USDA, and AgriLife cannot currently produce this verification data with any other system. That dataset grows in value with every unit deployed. No competitor can replicate it by buying a chip and writing code.
One DIN-rail mounted controller. A proprietary dataset that compounds. A patent on the trigger logic. That's the platform.
Active Member · Approved July 2026
NVIDIA Inception
Capital Connect
Sounder Solutions was accepted into NVIDIA's Inception program and approved for Inception Capital Connect (ICC) on July 29, 2026. Hundreds of venture capital firms in NVIDIA's ecosystem now have direct access to our profile and project data during the active 90-day window.
Active VC Visibility
Our company profile is live on NVIDIA's ICC platform through October 2026. NVIDIA's venture capital partners can review our technology, traction, and raise details directly.
Technical Validation
NVIDIA reviewed our Jetson Orin deployment, training pipeline, and technical roadmap before acceptance. Inception membership is their endorsement that this is a real AI company building on their platform.
Ecosystem Access
$10,000 in AWS GPU credits through Inception benefits, NVIDIA Deep Learning Institute certification, dedicated technical support, and Omniverse Replicator access for synthetic training data.
If you're here from ICC: You're looking at three hardware generations, a filed patent on the trigger logic, and an operational AWS training pipeline. The technology came before the business. That's why NVIDIA accepted us.
The hardware is built. The training is running.
Three hardware generations. A filed patent. An operational NVIDIA training pipeline. If you're here because NVIDIA sent you — the right move is a conversation, not a form.