AI
A step counter on a dog's collar sounds like it should be a solved problem. After all, human pedometers have shipped in phones and watches for over a decade. But a dog is not a human walking upright with a phone in their pocket. The sensor sits on
AI
At Hoomanely, a lot of our intelligence runs on models we trained in PyTorch: YOLO segmentation networks that understand what is happening in a frame, a MobileNetV3 classifier that recognises bowls, and a YAMNet-based audio model that listens for specific sounds. Training these in Python is comfortable. Shipping them is
AI
When teams build their first Retrieval-Augmented Generation (RAG) chatbot, most of the engineering effort goes into chunking, embeddings, and vector search. That is exactly where we started at Hoomanely. But after months of iterating on a pet-health assistant grounded in veterinary books, nutrition manuals, and clinical PDFs, we hit a
AI
How we achieved fully automated, production-grade image labeling by adding VLM validation as the final quality gate The Problem: The Last 15% of Labels Were Garbage We had built a sophisticated three-stage pipeline (YOLO → Grounding DINO → SAM3) that could generate thousands of segmentation labels automatically. But there was a problem:
AI
In a connected pet product, microphones can capture exactly what we need and a lot of what we do not. We want chewing, lapping, barking, and other pet-health signals. We do not want intelligible human conversation in stored audio. That is both a privacy concern and a trust issue. At
AI
We wanted centimeter-level alignment between our RGB camera and a low-resolution thermal sensor to measure eye temperature accurately. At 30–50 cm distance, small geometric errors translate into large thermal overlay shifts. A 3–4 cm depth misestimate can move the projected thermal blob several pixels — enough to shift from
AI
When an edge AI system starts to fail, it rarely fails all at once. Memory pressure creeps up. DMA stalls appear sporadically. Inference deadlines start slipping. Power draw spikes in ways the lab never showed. The question is not whether something will break in production — it’s what you choose
AI
Introduction Most edge AI pipelines are accidentally model-first. We optimize inference time, shave milliseconds off convolution layers, and debate INT8 versus INT4—while silently dropping sensor frames upstream. By the time the model runs, the data it sees is already compromised. On constrained hardware like STM32-class processors, this mistake is
AI
Most public audio datasets look clean on paper. Five seconds. Ten seconds. One label per clip. That assumption is exactly where our problems started. The Hidden Problem With Scraped Audio Datasets We started with the usual approach: scrape audio clips from public sources, bucket them by class (bark, howl, cough,
AI
When we added new audio classes to our MelCNN pipeline (sneezing, coughing, anomalies), something unintuitive happened: accuracy on existing, well-performing classes like eating and drinking dropped - even though their data hadn’t changed. At first glance, this looked like a training issue. It wasn’t. What Actually Broke 1.