AI Can See. AI Can Hear.
AI Has Never Felt.
Factories track everything — machines (MES), materials (ERP), warehouses (WMS). The one thing they don’t track: the 300 million workers on the floor. $500B lost every year to the last blind spot in manufacturing.
precision_manufacturingPerception Intelligence for Factory Operations
We Found It in the Wrist.
33 published datasets. 15 million recorded actions. 1.15 million participants across independent studies. One finding: temporal patterns in wrist acceleration encode behavioral states — fatigue, attention, skill proficiency, stress.
What AUC 0.737 means: out of every 10 judgments, the system correctly distinguishes fatigued from normal over 7 times. Camera systems: 0.60–0.68. We match their accuracy at $6 vs $200,000.
33 published datasets analyzed
14:15 — Zhang Wei, Plating Line B.
Fatigue score 0.73 (personal baseline 0.45). Plating rhythm slowed from 4.2s/board to 5.1s/board — 21% below his normal pace.
Verbal Reminder
paymentsZero costschedule5 secondsShift leader’s watch buzzes: ‘Plating B — Zhang Wei, pace dropped 21%.’ Leader walks over, quick check: ‘Watch the copper thickness.’
Target QC
paymentsZero costschedule30 secondsDownstream QC inspector gets notification: ‘Prioritize Plating Line B output, 14:00–14:30.’ Random sampling becomes targeted inspection. Each defect caught here saves ¥420. Caught at final test: ¥2,000+. Reaches customer: ¥5,000+.
Insert Break
paymentsLow costschedule15 minutesIf fatigue score hasn’t recovered after 15 minutes, system recommends a break. Shift leader decides. After break: copper thickness deviation returns from ±8.3μm to ±3.1μm.
These three actions exist in every factory today. Shift leaders already do them — by gut feel. We turn gut feel into data. 31 transferable decisions × 4 categories × 4 deployment phases.
See all 31 decisions →arrow_forwardA $5.82 Wristband That Looks Like a Factory ID.
Workers won’t refuse to wear it because it looks like what they already wear.
Layer 1: Dashboards — $3–5/worker/month
See what’s happening. Fatigue scores, attention levels, shift patterns.
Layer 2: Action Intelligence — $8–15/worker/month
Know what to do about it. The system searches 31 possible actions and tells you which one saves the most money right now.
← This is the product| Method | Cost | Accuracy |
|---|---|---|
Camera | $50K–200K/line | AUC 0.60–0.68 |
EEG | $200–500/unit | AUC 0.75–0.80 |
Self-report | ~$0 | Unreliable |
iFactory | $5.82 | AUC 0.737 |
We Didn’t Cold-Email Our Way Into Manufacturing.
My father has run a factory in Dongguan for 28 years. He makes caps for Hennessy, Chanel, and Dior. I grew up on that floor. This isn’t a thesis project — this is the problem I was born into.
trending_upThe Window
Edge AI chips hit $3 — nRF52832 runs inference in <50ms on a $2.40 chip
PIPL (2021) made cameras a legal liability — factories need an alternative
Labor costs tripled in a decade — the pain is acute now
Daizhe Zou
Founder & CEO
UC Berkeley, 19. Father’s factory: Wankun ihome Solutions, Dongguan, 28 years. F-1 visa (equity only).
Hongyu Xu
Co-Founder & CTO
UIUC CS + Neuroscience. ML pipeline architecture.
person_addNext hires: Embedded engineer, Time-series ML researcher, China BD lead
pin_dropBuilt at Berkeley. Tested in Dongguan.
Founded March 2026. Here’s What One Month Looks Like.
Company founded
Delaware C-Corp, 22.5M authorized shares
Research complete
9 internal papers, 33 published datasets analyzed
Patents drafted
11 USPTO provisionals, v7 final
$380K raised
Two SAFEs at $5M post-money cap (Family & Friends)
Product built
SDK AUC 0.737, FastAPI backend, hardware specs finalized
Pipeline active
Fushi (300852.SZ): proposal sent. Father’s factory: Phase 0 approved.
Key Metrics
Samsara proved that instrumenting a workforce’s tools creates a $20B company. We instrument the workforce itself.
$500K — $1M
Pre-money $5–10M · Already raised $380K via SAFE ($5M cap) · Monthly burn $3–5K · Runway 6+ years
Use of Funds
File Patent #1, Phase 0 pilot at father’s factory, register WFOE, hardware v1
Submit Paper A & B, Fushi Phase 1 deployment, 100-unit production
5–10 factory deployments, SDK licensing conversations, Series A prep
Your Money Buys:
Filed patents. Working hardware. Real factory data. First paying customers.
Berkeley, CA & Dongguan, China