ESP32-S3-based WiFi CSI platform detecting presence, vital signs, and fall detection through walls; pose estimation pipeline ships without trained weights.
Key Takeaways
Core sensing (breathing 6-30 BPM, heart rate 40-120 BPM, fall detection under 200ms, through-wall up to 5m) works today on ESP32-S3 ($9/node).
17-keypoint pose estimation infrastructure (WiFlow + AETHER + MERIDIAN heads) is implemented but pretrained .rvf weights are not yet released; skeleton display falls back to amplitude heuristics.
Presence detection and multi-person counting are heuristic/threshold-based, not learned classifiers; false positives under strong RF interference are a known issue.
Full BOM with Cognitum Seed (persistent vector store, Ed25519 witness chain, kNN) is $140; Docker image runs on simulated data with no hardware needed.
ESP32-C3 and original ESP32 are explicitly unsupported; two or more ESP32-S3 nodes recommended for meaningful spatial resolution.
Hacker News Comment Review
Skepticism centers on whether the repo is a vibe-coded promo project rather than a tested implementation; no independent hardware validation is cited.
One commenter points to commercially shipping WiFi sensing products (Kaspersky covered examples) as evidence the underlying physics is real, even if this specific repo is unverified.
Notable Comments
@rafram: “we have no way of knowing if the creator even bothered to test it with real ha[rdware]” – flags it as jargon-heavy vibe-code tied to a consulting business.
