Amit Patnaik Home
Sections

Amit Patnaik New Delhi

Communications professional. Public policy.
Software developer.

More than 15 years in corporate communications and HR, including experience working with senior government leadership. A self-taught developer, I use AI to build apps and tools of my own.

See the work My journey with AI

01 / 08 Work, writing, and my journey Scroll ↓

Peeku’s Mac interface showing the Desk Health view
PeekuMac · Desk health
FuelSaathi’s iPhone interface showing manufacturer confirmation and its source link
FuelSaathiiPhone · Manufacturer sources

02 / Selected work

The work

01 / Desktop app

Peeku

A desk-health app for long days on a computer.

Peeku offers eye-rest and movement reminders and a local record of your desk habits, with editions for Mac and Windows.

About Peeku ↗Visit Peeku ↗
Peeku’s Mac interface showing the Desk Health view
Mac Desk Health view, from Peeku’s public product images.

02 / For Indian drivers

FuelSaathi

E20 information for your car, with the source attached.

E20 is petrol blended with 20% ethanol. Choose a make, model, and year to look up the available guidance. FuelSaathi includes manufacturer references, mileage tools, and car-care advice, on the web and iPhone.

About FuelSaathi ↗Check your car ↗
FuelSaathi’s iPhone interface showing manufacturer confirmation and its source link
Manufacturer confirmation in FuelSaathi’s iPhone app, from its public App Store images.

Also live: MustEatList, Sift for X, Pace, and earlier work.

03 / My journey with AI

How I began building software

I started using AI to explore questions and work through drafts, then began learning to build software with it.

Learning by building

Peeku, my desk-health app, began with a problem I knew well: spending too long at the screen. FuelSaathi followed a question about E20 petrol and my car, where the answer needed a source I could check.

Once people could use the apps, I had to check the claims behind them and understand where users got stuck or needed a better explanation.

At my own desk

What I changed about my AI workspace

I kept repeating background, losing corrections, and getting plausible drafts in the wrong voice. I now keep relevant instructions and earlier decisions in an organised workspace and bring them into a task when needed.

04 / Behind the work

What is already in the model, and what I give it

Training shapes what a model can generally do. For a particular task, the answer also depends on which instructions, files, and tool results it can use.

01 / The model

General capability

A conversation does not permanently write a new fact into the model.

02 / The current task

What it can use now

Instructions, supplied material, and results from tools belong here. Missing material cannot reliably shape this answer.

03 / Saved records

Bring in what matters

A note saved elsewhere has to be retrieved. It may also be out of date.

Explore the AI stack ↗AI terms, explained ↗

05 / Making this site

How this site changed

Previous green and white homepage opening with top navigation and overlapping product views
Before: the previous homepage opening, captured locally on 24 September 2026.
Blue homepage draft with a left chapter rail and overlapping product views
Next draft: the blue opening on 25 September 2026. The headline and product layout changed again after this capture.

I first imagined this material as a presentation. An early site turned it into short screens and asked visitors to keep clicking Next. Using it showed me how little a visitor could see at once. I moved the work and the explanation onto one scrolling homepage, then kept the game near the end for anyone who wanted to try the idea.

06 / Play

Can you help Fetch finish the job?

Choose a dog and try the first task. If it gets stuck, change the setup and try again. Check what comes back before calling it a success.

Fetch a bottle

A pull strap makes the fridge usable. A routine or starting strength helps the dog complete the sequence.

Start a story

The relevant character note must reach the desk before the dog can use it.

Build a small website

An open route, usable workbench, and working preview all matter.

The interactive game loads when you reach this section. The examples above remain the explanation if scripts are unavailable.

A routine in Fetch stands for reusable guidance, not permanent model training. The game’s rules are fixed; real AI results are less predictable.

07 / Selected writing

Writing

Essay / AI in practice

The configuration layer

Why I think the practical AI opportunity for Indian institutions lies in fitting existing models to their work.

Read the essay ↗
Explainer / AI infrastructure

The AI stack

Follow an answer from the app down through serving, models, chips, and power.

Explore the stack ↗
Essay / Communication and software

Make the system legible

What writing an explanation and designing a useful interface have in common.

Read the essay ↗
Browse all writing

08 / About

Get in touch

I’m based in New Delhi. My apps and tools are independent projects, alongside my work in communications and public policy.

Email me ↗LinkedIn ↗More about me ↗