Skip to content
Raj Bhensdadiya

Quellit AI

It reads the requirements a team already wrote, turns them into end-to-end tests, and runs them in a real browser.

Role
Full-Stack & AI Engineer
Timeline
7 months
Live site
Visit the live site
JiraGitHubFigma20 agentsTest casesCloud browserrun and verifyCredit ledgerquota, refundsFailed run, refunded
How it works: requirements from Jira, GitHub and Figma go to 20 CrewAI agents that write test cases and drive a real cloud browser to run them, while a credit ledger meters each run and refunds a failed one.
LLM agents
20
third-party integrations
12
metered quota entry points
11

Overview

Quellit AI runs end-to-end web QA with a multi-agent engine. It reads a team's Jira, GitHub, and Figma requirements, writes the test cases from them, and then drives a real cloud browser to execute and verify each one.

The problem

Writing end-to-end tests by hand is slow, and they rot the moment the UI moves. What teams wanted was cases generated from the requirements they already write, then run against the real product rather than a mock.

What I built

Built a CrewAI multi-agent engine coordinating 20 LLM agents that turn Jira, GitHub, and Figma requirements into test cases, then drive a real cloud browser to execute and verify them. A usage-based credit ledger meters every AI test run and enforces quota at 11 entry points, refunding automatically when a run fails, on top of Stripe subscriptions and pay-as-you-go top-ups.

Results

Integrated 12 third-party systems, including a signature-verified GitHub webhook that turns every opened pull request into an AI-generated test plan. Linear, Vercel, and Figma OAuth are brokered through Nango.

Decisions

  • Start from requirements teams already write

    Test cases come from the Jira, GitHub and Figma requirements a team already has, so nobody writes end-to-end tests by hand.

  • Run against the real product

    Agents drive a real cloud browser rather than a mock, so a test fails when the product does.

  • Meter every run, refund the failures

    A credit ledger enforces quota at 11 entry points and refunds automatically when a run fails.

Screenshot of the Quell product site, showing its tool integrations for GitHub, Linear, Figma, Slack, Vercel and Jira
The live product site.

Stack

  • Next.js
  • TypeScript
  • Node.js
  • FastAPI
  • PostgreSQL
  • CrewAI
  • OpenAI
  • BrowserBase
  • Stagehand
  • Stripe
  • Nango
  • GCP

Built by Raj Bhensdadiya, an AI & full-stack engineer in Montreal. The full work history and the other projects have more, and email is the fastest way to reach me.

More work