Nisarg GurjarEmail me
Founder of Ajna · four products in production

I direct AI to build software. My job is knowing when it’s wrong.

I’m Nisarg Gurjar. I’m a founder and operator. My team is AI; my job is judgment — what to build, what’s correct, what ships. The result: live SaaS, automation carrying real revenue for a real events company, an iOS app, a desktop app. Everything in this story is real, tested, and in use.

4products live
2Slack bots in daily use
670+automated tests, all green
1iOS app on TestFlight
1notarized Mac app
SCROLL — OR USE THE MAP →
Scene 01LIVE · MY VENTURE

Ajna. A sales call becomes an approved quote, for any event vendor.

Multi-tenant SaaS: call transcript in → reviewed, human-approved, branded quote email out.

Days → minutesquote turnaround, with a human always in the loop
Open the case study
Quote · Riverside receptionREADY FOR REVIEW
Transcript parsed · 14 fields98% CONF
Branded email previewRENDERED
Scene 02IN PRODUCTION · M4U EVENTS

Quote Automation. Real revenue, through a pipeline the AI can’t misuse.

A sales call becomes a branded email, calendar invite, CRM log, and Slack update — after a human taps Approve.

480 tests + an adversarial eval gatestand between the model and a client’s inbox
Open the case study
Transcript ready · wedding inquiry · extraction complete
✓ Approve & send
Sent. Calendar invite created · CRM logged · thread updated
Scene 03LIVE · ~30 USERS

Workflow Tracker. Hundreds of event tasks, zero silently dropped.

Task tracking for an events company — a workflow engine with prerequisites and triggers, plus a Slack bot that escalates overdue work.

~30 people, every event, every weekrun their checklists through it
Open the case study
Load-in checklist · SatON TRACK
Sound check · blocked by 2TRIGGERED
Photo proof · uploadedVERIFIED
Scene 04 — Everything else that shipped
Scene 05 — Method

I decide what should exist. AI writes the code. A system of gates decides what ships.

METHOD 01My own AI tooling

12 custom skills + 6 guardrail hooks — including guardrails against my own AI touching secrets or sending real email.

METHOD 02Eval gates

LLM features pass adversarial, must-be-100% eval suites before going live. Twice, they caught what tests couldn’t.

METHOD 03Shipped means walked

Nothing is done until the real user flow works end to end — logs checked, database verified, success seen.

The full method — how I ship tested software with an AI team

Run an events business? I built Ajna for you.

AI automation for wedding & event vendors — quotes in minutes instead of days, a human always in the loop.

Visit ajnadesk.com