Production AI for mission-critical operations · San Jose, CA
We build production-grade voice, knowledge and operations AI for industries where downtime is dangerous and expensive. Proven with KONE — a Fortune 500 Europe elevator company — and expanding from elevators to enterprises.
Simulation · scripted from the production call flow
Outcomes delivered for KONE
2.3 s
time to first audio (was 6.8 s)
−30%
emergency intervention time
90%
grounded answer accuracy (up from 80%)
~$1M
client cost savings, one programme
Clients · Partners · Pedigree
Advised by former VPs & SVPs of KONE and TK Elevator.
The problem
Traditional service operations still run on manual call handling, fragmented reporting and slow technician response. The result is downtime, cost and frightened passengers.
30+min
average rescue response today
1queue
a trapped passenger waits behind everyone else at peak
0shared context
the operator, the dispatcher and the technician each start from scratch
Rescue response
20M+
elevators and escalators in service worldwide — and the same pattern (installed equipment, field technicians, emergency calls, legacy dispatch) repeats across HVAC, building automation, industrial equipment, fleets and medical devices. Industry estimates
Platform
Voice, knowledge, operations and vision — reusable building blocks for any equipment-heavy operation, integrated with the systems you already run.
01 · Voice
Emergency & call-centre automation. Streaming speech recognition → an LLM grounded in your SOPs → streaming synthesis, with barge-in and a four-tier fallback that ends with a human. Multilingual.
02 · Knowledge
Sales, service and maintenance assistants that answer only from your documents: fault-code triage, procedures, pricing, policy. 80% → 90% accuracy, with a feedback loop from real query logs.
03 · Operations
Tickets, technician dispatch, predictive scheduling and a real-time uptime dashboard — integrated with your CRM, PLCs and dispatch systems instead of replacing them.
04 · Vision
Elevator interior concept design for sales teams; batch visual generation on self-hosted models with GPU scheduling; computer-vision inspection (vehicle damage detection) for mobility.
Inside one emergency call
Scroll through the production architecture: the path an emergency call takes from a stalled car to a spoken answer — and what happens when something fails.
Someone is trapped on floor 12. The call hits Twilio and streams over a WebSocket to a stateless worker. No keypad menu, no queue.
Azure streaming recognition with tuned endpointing turns speech into text while the passenger is still talking. Barge-in means they can interrupt.
The LLM answers from the emergency SOPs via RAG. Every piece of call state lives in Redis under the CallSid, so any worker can take over. A watchdog reassigns a call ten seconds after a heartbeat goes quiet.
Streaming synthesis overlaps with generation. If a stage times out: shorter prompt → cached answer → human operator with full context. The caller never hears silence.
One conversational turn
~3×faster to first audio
seconds after the caller stops speaking
Stage boundaries are illustrative; first-audio times are typical production values from delivery.
Proof
Requirements, architecture, launch and operations — owned end to end.
Enterprise · KONE
Instant, multilingual emergency response for trapped passengers.
Enterprise · KONE
Ten seconds in front of a client to answer model, configuration, price and policy — from hundreds of PDFs.
Enterprise · KONE
Instant troubleshooting and proactive support for field technicians.
Enterprise · KONE
Turn “make it feel more premium” into a rendering the client can react to.
“Rent the Future of Driving.”
An AI-powered car-rental platform focused on advanced driver-assistance vehicles: computer-vision damage inspection, fleet intelligence, dynamic pricing.
Figures from project delivery; client internal estimates; not independently audited.
Why we win
~30%the model
~70%state · failover · scaling · permissions · migrations · observability
Grounded in your SOPs, fault codes and manuals — it answers only from your documents.
State in Redis, heartbeat watchdogs and a four-tier fallback from day one.
Connects to your CRM, PLCs and dispatch systems instead of replacing them.
AI on Azure for better Mandarin speech, infrastructure on AWS — ~100–300 ms per call, accepted for voice quality.
Four programmes shipped for KONE, each with a measured outcome.
Guided by former VPs and SVPs of leading elevator companies.
Traction & roadmap
From one elevator company's hardest call to every industry that runs on installed equipment.
First KONE engagement.
Sales assistant, maintenance knowledge, interior design, emergency voice.
DriveWe mobility platform.
“From elevators to enterprises.”
Team
Co-founder & CEO
Runs the company, customer relationships and the mobility vertical. Stanford MS&E (Technology Ventures Program); M.Eng. EECS, UC Berkeley; B.S. Electrical Engineering & Cognitive Science, UC San Diego. Previously AI/ML solution architect at Dell Technologies.
Co-founder & CTO
Led all four KONE AI programmes end to end — requirements → architecture → launch → operations. M.S. Machine Learning & Data Science, UC San Diego.
Also founder & CEO of DineCall AI, a restaurant AI phone platform in use at multiple Vancouver restaurants with clear efficiency gains — proof of production discipline:
Recognition Silver, 6th China International “Internet+” Innovation & Entrepreneurship Competition (2020) · Gold, Microsoft Imagine Cup (2019)
Team trained at Stanford, UC Berkeley, USC and UC San Diego · Advised by former VPs & SVPs at KONE and TK Elevator.
Plus AI engineers and a network-security specialist.
What clients say
The AI Chatbot created by We Link Technology has completely captured our company's essence. Their exceptional service surpassed all our expectations — we couldn't be happier.
For investors
Voice latency is finally solved: 2.3 s end-to-end.
LLM + RAG makes decades of manuals usable.
Legacy service industries face labour shortages and rising uptime SLAs.
A Fortune 500 reference customer and four shipped programmes with measured outcomes.
A reusable platform: voice, knowledge, operations, vision.
A second vertical already live.
Contact
Tell us about your operation — the equipment, the call flows, the systems you already run — and we'll show you what production AI looks like on it.
Phone
(650) 776 5335Locations
San Jose, CA (HQ) · Vancouver, BC · Austin, TX
