Our work

Systems we built, still running.

What we designed, built and operate today, and what each one does: for Preeco in Srinagar, for clinics as our own product, for the Kashmiri language, and this website. Told plainly.

Preeco

Beauty and skincare retail · Srinagar

A shopping assistant that knows the shop

The challenge
Connect shopping conversations on WhatsApp, in four languages, to the store’s current catalogue, stock and prices, with a person able to step in.
The system
A WhatsApp assistant connected to the store’s own catalogue, stock and prices. It reads Kashmiri, Urdu, Hindi and English, in script or romanised, recommends products with the shop’s care advice, and hands over to staff with the thread intact.
How it works
  1. A message arrives and its language and intent are read
  2. The store’s data is checked before a word is said
  3. The answer comes back as text and product cards
  4. Anything it should not decide goes to a person
  5. Real conversations feed improvements the owner approves
What people use
Customers get answers in their own language. Staff pick up handovers with the whole conversation. The owner accepts, rejects or trials each proposed improvement from his phone.
What it does now
Shopping answers come from the store’s data, consent is asked for and recorded, and every change to the assistant is recorded with a way back.
Built with
WhatsApp · Workflow automation (n8n) · PostgreSQL · The store’s API · A local Kashmiri language model · Hosted models under a budget
Our role
Designed, built and operated by Omvora. See Chatbots.

Retail operations · Srinagar

Orders and a nightly view of the business

The challenge
Give staff one place to move orders through their stages, and give the owner one view of what sold, what customers wanted and what was missing.
The system
Order operations on Telegram, with each order moving through its stages and roles for staff, and a report to the owner every night.
How it works
  1. An order enters and is placed at its stage
  2. Staff move it on from Telegram
  3. Each night the day is summarised: stock, searches, sales
  4. A written summary is added by a local model
What people use
Staff work where they already are, in Telegram. The owner reads one page each night: stock by category, what people searched for and did not find, products tapped that cannot be sold, paid and started checkouts.
What it does now
The team follows orders stage by stage in one place, and the owner sees each evening where demand is going unmet.
See a sample
A redacted copy of Preeco’s nightly insights report as it arrives on Telegram: stock by category, what people asked for, card taps, money, the assistant’s quality scores and what to keep doing. Every figure, product, brand and customer’s word is replaced by a dark bar.
The owner’s nightly report, redacted: every figure, product, brand and customer’s word is a bar; only the report’s own wording and the store’s public categories remain.
Built with
Telegram · Workflow automation (n8n) · PostgreSQL · A local language model
Our role
Designed, built and operated by Omvora. See Digital products · Chatbots.

Social automation and content · Srinagar

From store footage to approved posts

The challenge
Turn real store footage and planned content into posts through a controlled review and publishing process.
The system
A content calendar with approval cards on Telegram and scheduled publishing, and a video studio that turns real store footage into reels.
How it works
  1. Footage is analysed shot by shot: cuts, speech, what is on screen
  2. A local model directs the edit to a brief
  3. The reel is rendered and checked automatically
  4. An approval card reaches the owner; approved work is queued
What people use
The owner films, approves and rejects. Everything else, from the plan to the cut to the schedule, runs as a system.
What it does now
Content keeps going out on a plan, made from the shop’s own footage, and nothing is published without approval.
Built with
Instagram and Facebook · Telegram (approvals) · Scene detection and speech recognition · A local vision and language model · FFmpeg
Our role
Designed, built and operated by Omvora. See Social automation · Idea to post.

Our own product

SaaS for private clinics · Kashmir and Ladakh

Omvora Clinic Intelligence

The challenge
Bring bookings, queues, reminders and payments into one administrative workflow, with each clinic’s records kept apart.
The system
A product run entirely inside Telegram: bookings, live token queues, reminders and payments. Administrative only, never clinical.
How it works
  1. A patient books from a link in Telegram
  2. Consent is recorded for each purpose
  3. The queue is live; the patient sees their token
  4. Reminders go out; payments are taken for the clinic
What people use
Patients use the app they already have. The front desk works from one screen. Each clinic’s data is kept apart at the database layer.
What it does now
Built so patients book and follow their token from their phone, with an audit record of the workflow’s actions.
Built with
Telegram · PostgreSQL with row-level security · Workflow automation (n8n) · Online payments
Our role
Our own product: designed, built and operated by Omvora. See SaaS.

Our own work

Language technology · Srinagar

A language model for Kashmiri

The challenge
Understand customers who write in Kashmiri, often romanised, and give native speakers a way to correct the model.
The system
A local 27-billion-parameter model adapted for Kashmiri, with a review app in which native speakers correct its output round by round.
How it works
  1. Native readers review and correct its answers
  2. Corrections become the next round’s training data
  3. The model serves customer conversations from our own servers
What people use
A customer writes the way they speak and is answered in kind.
What it does now
A Kashmiri language capability that runs on our own servers, with native speakers correcting it round by round.
Built with
An open 27B model · Fine-tuning (LoRA) · Our own GPU server · A native-speaker review app
Our role
Built and run by Omvora. See Chatbots.

Omvora

Digital product · omvoramedia.com

This website

The challenge
Explain what Omvora builds through a website that is itself the evidence: clear, fast, accessible and honest about data.
The system
A static site with a scroll-driven story, a text engine that reveals every paragraph with the scroll, a liquid-glass interface, and consent receipts for every choice a visitor makes.
How it works
  1. Built as static pages, served from our server with a Cloudflare mirror
  2. Every page audited against WCAG 2.2 AA, with a published conformance report
  3. Consent recorded as a receipt, kept for a year
  4. A security gate before every release
What people use
Visitors read, scroll and choose; people who prefer less motion turn it off.
What it does now
A site audited against WCAG 2.2 AA, with a published conformance report, and a consent receipt for every visitor’s choice.
Built with
Astro · GSAP · nginx · Cloudflare · Playwright test harnesses
Our role
Designed, built and operated by Omvora. See Digital products.

Have something that should run like this?

Tell us what happens today and what you wish happened. We will tell you what a system could take off your hands, and what it should not.

Talk to usHow we work