Jobs today
186
We wire models into the work you already run — WhatsApp, invoices, PMS, POS, and CRM — with audit logs, human gates, and Kathmandu delivery. No stock photos. No mystery black box.
12+
Live workflows
< 8s
Typical handoff
Human
In the loop
4–10 wk
First production run
Generic AI pages sell a prompt box. Operators in Nepal need a closed loop: an event, a decision, a write-back, and a person who can stop it. That is what we ship — gold-and-ink chrome, dense data, motion that does not hang the scroll.
A booking, a WhatsApp ping, an overdue invoice, a low-stock SKU — not a weekly CSV dump someone forgot to attach.
Classify, extract, draft, or score. The model proposes; your policy decides what can fire without a person.
PMS, billing, WMS, CRM. If it does not land in the same ledger your floor already trusts, it is a demo — not automation.
WhatsApp and web chat that know room type, arrival, and balance — then hand off to a person when the guest is angry or the request is odd.
PDF and photo invoices into GST-ready lines. A clerk confirms the exception pile; the rest posts.
Night audit notes, walk-in quotes, and housekeeping flags drafted from the PMS — not a second spreadsheet.
Reorder drafts from bin levels and vendor lead times. Purchasing still hits approve.
CRM tasks from stalled deals, with a draft in the owner’s voice and a reason the model used.
Ticket routing by product, language, and severity. Finance never sees a password-reset thread again.
First drafts for trek departures and room copy, with a human editor and a brand sheet — not unreviewed blog sludge.
Staff search over your runbooks and SOPs. Answers cite the paragraph. No hallucinated policy.
Refunds, rate changes, and guest-facing copy above a threshold wait for a named role.
You choose the model host. We do not train public weights on your folios or invoices.
Every action can be replayed: prompt version, tool call, and the row it wrote.
One toggle pauses a workflow without taking the rest of the product offline.
We sit with the people who already do the work and list the ten jobs that waste the most hours.
What the model may do alone, what needs a person, and what is never automated.
The workflow proposes; staff still click. We measure precision before it writes.
Production queue, alerts, and a named engineer for the first thirty days.
Tell us the system, the trigger, and who must approve. We reply with a workflow plan — not a generic AI brochure.