Engineering AI
Research, product updates, and deep dives into ML for engineering assessment, PCB design scoring, and the SRHN reasoning engine.
Traditional ML confidence is empirical — "our model said 82%". PAC-Bayes provides theoretical guarantees derived from spectral norms of the knowledge graph. Here's the difference in practice.
Multi-semantic orthogonal views, hyperedge attention, temporal edge decay, and Laplacian spectral reranking — how we built a C11 reasoning engine from first principles.
An engineer who used the trace width calculator before submitting scores higher on signal integrity. This isn't a reward — it's a genuine, hard-to-fake quality signal.
PCB geometry + schematic + SPICE simulation + MCAD + interaction trace + SRHN reasoning — why a 192-dim vector beats any single-modality approach to engineering scoring.
68% of tech interviews test algorithms. Only 4% include any evaluation of actual design work. This is the gap EngineIQ was built to close — for PCB engineers, analog designers, and embedded systems developers.
We trained a 6-layer Graph Attention Network on KiCad schematics treated as component-net graphs. Here is what the learned 256-dimensional embeddings capture — and what they miss.
Stay updated
One email per post. No marketing. Unsubscribe any time.
Subscribe via email →