A design studio that closes the gap between idea, design and code

In every company I worked in, I saw the same gap. Founders had good ideas. Designers designed them well. Developers built them. But the three handoffs leaked: ideas reached designers without validation, and what got built was not what was designed. A good idea with top-notch execution still fell apart in between.

So I put a designer into ideation, to do user and market research, validate the idea and shape an MVP before a startup spends time and money. The name comes from the Marathi word for a kite: we take your idea and make it fly.

1. 1

   Deep-dive desk research on the domain, the requirements and the users: what they want, what they need, and whether the real problem is being solved.

2. 2

   Mood board for visual direction: imagery, colour, typography, icons.

3. 3

   Scope a small MVP instead of building a whole product, and test it with a few target users.

4. 4

   Revise the MVP, whether it was validated or not, then hand it to development.

5. 5

   Hand over code and design to the founders.

6. 6

   Support new features for one year. After that, further work is a new project.

A learning management and operations platform for schools, in use at two schools in Pune by 35+ teachers (about 70 people with admins and content managers). Three roles: admin (timetable, slots, teachers), content manager (adds lesson content with AI) and teacher (follows the timetable, reports period status and findings).

The insight: the founder brought the idea. We interviewed teachers, admins and principals, and I taught as a guest teacher in some lectures to see a day first-hand. We learned that education staff are busy and not eager to adopt new tools. Teachers love teaching. Administration is what they have to do.

The decision: I separated administration from teaching, so teachers see only what they need to teach and report.

What failed: we tested with 7 teachers. The core flows worked. The timetable algorithm did not. It assumed an ideal school and broke on edge cases, such as a teacher leaving mid-semester, a teacher being absent, or last-minute changes. We fixed it by adding a reinforcement learning algorithm with a manual override.

The gap: AI calling was aimed at the US and big banks. Small and medium businesses, especially real estate, insurance and small marketing firms, were not reached.

What we did: competitor research and conversations with real users, then the name, logo and brand, then an MVP tested with a few businesses. That showed three use cases: real estate, insurance agents and collections. We built an add-on for each. We maintain the server infrastructure, update the voice models and have added languages.

Result: used by three organisations, about 50 to 60 users.

The client wanted an AI learning layer for an existing product that was supported by customer support, demos and guides. They sent their design language: icons, fonts, colours.

Our idea: a podcast-style audio library of short how-to guides, created with AI and personalised to each user's goals and personality. We designed three AI agents, each with its own personality and expertise, each building a library, shared across users on request.

Scope: design only. The client's developers built it. We went from requirements to the full design and handoff.

I built a system inside the studio that takes an MVP from idea to execution using 24 specialised agents, each with a defined task. It has produced a CRM and an automated collection platform for a client, and our own tools: a leads generator, a CRM, a chat application and an invoice generator.

A privacy-first work-intelligence product, in pilot with four organisations, about 40 users, capped at 10 per organisation per test. I wrote the core algorithm for capturing work patterns. It has its own page.

I am strong at design and development, and I underestimated how much business, marketing and winning clients would matter. Learning that has been the hardest part of running the studio.