# Huzuri

AI SaaS, 2026.
Live site: https://huzuri.in

Our own AI office app: attendance, follow ups and minutes, on autopilot.

## Objective

In most small offices the owner is the follow up system: checking who came in, chasing tasks by phone, remembering what was decided. We set out to build software that does the chasing itself, works with hardware the office already has, and brings people only the decisions.

## Approach

- Six jobs share one record of who is here, what is due and what was agreed.
- People keep the decisions. The software ranks, nudges and drafts; a person approves.
- Every output carries its evidence: a snapshot for each attendance mark, a line from the CV for each score, a clip for each decision.
- Nothing new to install: the browser, existing CCTV cameras, email and calendar.
- The marketing site tells the story as one simulated office day, and is written for search engines and AI assistants as well as people.

## How the system fits together

Parts:
- Cameras (build layer, channel)
- Attendance (intelligence layer, engine)
- Payroll (intelligence layer, engine)
- Meetings (build layer, surface)
- Minutes (intelligence layer, engine)
- Email (intelligence layer, engine)
- CV screening (intelligence layer, engine)
- Tasks (intelligence layer, store)
- Nudges (intelligence layer, engine)
- Assistant (intelligence layer, engine)
- Apps (build layer, surface)
- Website (growth layer, surface)
- Search (growth layer, channel)

Connections:
- Cameras → Attendance
- Attendance → Payroll
- Meetings → Minutes
- Minutes → Tasks
- Email → Tasks
- CV screening → Tasks
- Tasks → Nudges
- Nudges → Apps
- Tasks → Assistant
- Assistant → Apps
- Website → Apps
- Search → Website

## Features

### Attendance from existing cameras

Office CCTV marks attendance with a liveness check, so a photo held up to the camera doesn't count. Video stays on site, and payroll comes from the same record.

### Tasks that chase themselves

Every task has an owner and a date. Huzuri nudges, then escalates to the manager, then asks for approval.

### Meetings that end with tasks

Hindi, English and mixed conversations become minutes, and action items become owned tasks.

### An assistant that respects roles

Ask what's overdue or who's on leave. Answers cite their sources, and the assistant confirms before it acts.

### CV screening with reasons

Each CV is scored against the job description, with the line that justifies every score. A person makes the shortlist.

## Craft

- About 98 KB of code on the homepage and no third party requests.
- Product videos load only when needed; the launch film loads only when asked for, with captions.
- Full reduced motion support, one H1 per page, canonical URLs and FAQ schema on every page.
- An llms.txt file and an explicit policy that lets AI search crawlers read the site.

## Results

Founding pilots are being set up. We'll publish results once they're measured.

## Credits

Credits: the team. Disciplines: Product design, Web, Applied AI, Automation, SEO & AI search.
Stack: Next.js 16, React 19, Tailwind CSS, GSAP + ScrollTrigger, Google sign in, REST API.

Source: https://magrus.vercel.app/work/huzuri
