Back to home

All Writing

Essays and build logs about products, AI, philosophy, ambition, and figuring things out while I’m still in the middle of it.

IPMAT

Build notes from the IPMAT preparation ecosystem I am developing—beginning with Verbal Ability and growing toward Quant, exam practice and shared preparation intelligence.

IPMAT•Sep 17, 2026•7 min read

Why I Built My Own IPMAT Vocab App

I had too much vocabulary to learn for IPMAT, and the tools I was using never quite matched the way I wanted to practise. So I started building my own.

Read full article
IPMAT•Sep 19, 2026•7 min read

The First Version Had Only 32 Questions

After all the research, I finally stopped planning and built the first usable version of my IPMAT verbal practice app. It had 32 questions, browser storage, short sessions, and almost nothing else.

Read full article
IPMAT•Sep 22, 2026•11 min read

I'm Building This Until It Becomes the IPMAT App I Actually Want

What started as a tiny tool for my own vocabulary backlog now has practice, review, vocabulary learning, offline progress and sync. The next stage is about turning all of that into the exam-prep app I genuinely want to use every day.

Read full article
IPMAT•Sep 23, 2026•11 min read

I Stopped Building a Vocabulary List and Built a Vocabulary Engine

The vocabulary section had grown far beyond the 154 cards I started with. So I stopped treating vocabulary as content to browse and started building an engine that could decide what I should learn, review, and repair.

Read full article
IPMAT•Sep 24, 2026•11 min read

I Finally Built the Exam Mode I Kept Talking About

For months, I kept saying the app would eventually separate normal learning from strict exam simulation. After finishing the vocabulary engine, I finally built the exam engine properly.

Read full article
IPMAT•Sep 25, 2026•11 min read

My App Started Telling Me What I Was Bad At

The app already knew what I answered, what I got wrong, how long I took, and what kept coming back. The next step was making it interpret that evidence instead of dumping charts on me.

Read full article
IPMAT•Sep 26, 2026•10 min read

The Boring Features That Decide Whether I Can Trust My Own App

After building vocabulary intelligence, exam simulation and personal analytics, the next milestone was less exciting but more important: making sure progress survives bad internet, updates, sync conflicts, device changes and account actions.

Read full article
IPMAT•Sep 28, 2026•12 min read

I Built an App Behind My App

The student-facing app was becoming more capable, but managing its questions and learning content through raw files was becoming a problem of its own. So I built a private Content Studio behind the product.

Read full article
IPMAT•Sep 29, 2026•15 min read

Four Options Were Making Me Feel Smarter Than I Was

My vocabulary engine could tell when I picked the right answer. It still couldn't tell whether I could recall, distinguish, understand and retain a word. So I stopped treating one correct MCQ as proof that I knew it.

Read full article
IPMAT•Sep 30, 2026•12 min read

I Put AI in the App—and Refused to Let It Decide What I Know

I finally found a place where AI could genuinely help the vocabulary system: giving feedback on open-ended sentence practice. But I did not want a language model deciding whether I had mastered a word.

Read full article
IPMAT•Oct 1, 2026•11 min read

Then I Closed the Tab Mid-Exam

The exam engine worked beautifully as long as nothing interrupted it. Then I started thinking about refreshes, closed tabs, crashes and unfinished attempts—and realised an active session needed to survive the real world too.

Read full article
IPMAT•Oct 2, 2026•15 min read

It Worked for Me. That Wasn’t Good Enough Anymore

Once the app had accounts, synced progress, exam sessions and personal history, a bug stopped being something I could simply tolerate. I started hardening the product for the possibility that someone other than me might trust it.

Read full article
IPMAT•Oct 3, 2026•14 min read

This Started With 32 Questions. Here’s What It Became

The first version was a tiny browser app with 32 questions. It now has vocabulary learning, exam simulation, personal intelligence, accounts, sync, recovery and an entire content system behind it—but I still wouldn't call it finished.

Read full article
IPMAT•Oct 4, 2026•7 min read

My Verbal App Turned Into an IPMAT Ecosystem

I started by building one vocabulary app for myself. Then Verbal became a serious learning product, Quant became its own app, and I realised the interesting problem was no longer one subject—it was understanding the preparation happening across both.

Read full article
IPMAT•Oct 5, 2026•9 min read

I Built the Quant App I Wish I Had for IPMAT

Building Verbal taught me how quickly a study app can become shallow, fragile or dishonest. So when I started Quant, I approached it differently: syllabus first, evidence first, and a learning loop designed around how quantitative aptitude actually behaves.

Read full article
IPMAT•Oct 6, 2026•9 min read

I Didn’t Want Another Dashboard Full of Numbers

Accuracy, time and question counts are easy to measure. Understanding what they actually mean is much harder. Building Quant analytics became an exercise in learning when a number is evidence—and when it is just decoration.

Read full article
IPMAT•Oct 8, 2026•7 min read

The Quant App Finally Started Feeling Like a Real Product

The Quant engine had become capable long before the interface felt finished. I rebuilt the learner experience around focus, hierarchy, visual evidence and a calmer path from opening the app to actually practising.

Read full article
IPMAT•Oct 9, 2026•7 min read

EBO Died. Its Best Idea Didn’t.

EBO failed because understanding studying required too much manual context or too much observation. Verbal and Quant changed that: the evidence now appears naturally while I study.

Read full article
IPMAT•Oct 10, 2026•10 min read

Verbal Knows Verbal. Quant Knows Quant. Intelligence Knows Me.

Once Verbal and Quant both became serious products, I needed somewhere to understand the learner across them. That became Intelligence: a separate system for synthesis, planning, retention, mocks and evidence-backed next actions.

Read full article
IPMAT•Oct 11, 2026•9 min read

I Taught My App the Difference Between Evidence and a Guess

An intelligent learning system can become dangerous the moment it presents inference as fact. So I built the learner model around observations, confidence, contradictions, versioned hypotheses and the right to say 'not enough evidence.'

Read full article
IPMAT•Oct 12, 2026•8 min read

Three Apps Needed One Memory

Verbal, Quant and Intelligence could not become one ecosystem if each remembered the learner differently. So I started building a shared evidence platform with stable events, one identity, retries, versioned state and a common language for learning.

Read full article
IPMAT•Oct 13, 2026•9 min read

I Put My Web App on Android and Immediately Hated It

Getting the Verbal app to run on Android was much easier than making it feel like an Android app. Safe areas, back behaviour, haptics, keyboards, recovery and release engineering exposed the difference between packaging a website and building a mobile product.

Read full article
IPMAT•Oct 14, 2026•10 min read

Where the IPMAT Ecosystem Actually Stands Now

The project that began as 32 Verbal questions now contains a deep Verbal product, a serious Quant architecture, a separate Intelligence system, a shared evidence platform and active native work. It is much larger—and still much less finished—than the screenshots suggest.

Read full article

SYNTHESIS

Build notes documenting the development of my local AI assistant.

SYNTHESIS•Foundation · Part 1•8 min read

I Wanted to Build My Own JARVIS

SYNTHESIS began with an embarrassingly ambitious idea: I wanted a personal AI assistant that could understand my world, use tools, and actually do things—not merely answer questions.

Read full article
SYNTHESIS•Foundation · Part 2•8 min read

A Chatbot Wasn’t Enough

SYNTHESIS could not become the assistant I imagined by generating better replies alone. It needed context, memory, tools, permissions, execution, and an interface built around action.

Read full article
SYNTHESIS•Foundation · Part 3•8 min read

The First Time SYNTHESIS Actually Did Something

The first working version of SYNTHESIS had no voice, cinematic interface, or real intelligence. It could register two tools, judge their risk, execute one, and refuse the other—and that was enough to prove the foundation.

Read full article
SYNTHESIS•Foundation · Part 4•8 min read

Power Needs Permission

An AI assistant becomes useful when it can act. It becomes dangerous when action is confused with authority. SYNTHESIS is teaching me that capability must grow alongside permission, explanation, and verification.

Read full article
SYNTHESIS•Foundation · Part 5•7 min read

Teaching SYNTHESIS to Notice Its Environment

After SYNTHESIS learned to register tools and respect permission boundaries, the next step was giving it grounded context about the computer on which it was running.

Read full article
SYNTHESIS•Foundation · Part 6•8 min read

A Working Demo Is Not a Dependable Assistant

SYNTHESIS had passed its first tool, permission, model, and context tests. That proved its architecture could work—but a successful demonstration is very different from an assistant I can trust every day.

Read full article
SYNTHESIS•Foundation · Part 7•8 min read

Where SYNTHESIS Actually Stands

SYNTHESIS now has tested foundations for models, tools, permissions, execution, verification, and system context. Here is what genuinely works, what remains limited, and what comes next.

Read full article

EBO

The complete story of a student-accountability product I researched, prototyped and ultimately chose to end as a standalone product.

EBO•Foundation · Part 1•7 min read

EBO Was Never Supposed to Be a Product

EBO began as a small tool I wanted for myself. This is how a personal attempt to study more consistently slowly became a much larger product question.

Read full article
EBO•Foundation · Part 2•6 min read

When More Features Felt Like Progress

The early version of EBO kept growing—trackers, points, rewards, achievements, themes, and more. It took me time to realise that adding features was not the same as building a product.

Read full article
EBO•Foundation · Part 3•7 min read

The Questions That Changed EBO

EBO began changing when I stopped asking what else I could add and started asking whether the product deserved to exist at all.

Read full article
EBO•Foundation · Part 4•7 min read

Designing Accountability Without Surveillance

Accountability can help students remain consistent, but visibility can easily become pressure or surveillance. EBO forced me to think seriously about where that boundary belongs.

Read full article
EBO•Foundation · Part 5•7 min read

How I’m Researching EBO Before Pretending It Works

Believing that students struggle with consistency does not prove that EBO is the right solution. I’m learning how to search for evidence without turning research into reassurance.

Read full article
EBO•Sep 17, 2026•8 min read

I Finally Made EBO Look Like the Product in My Head

For months, EBO mostly existed as ideas, documents, rough builds, and conversations. I finally built a proper product experience that shows what I’ve actually been trying to make.

Read full article
EBO•Sep 19, 2026•8 min read

My First Student Interviews Challenged Some of My EBO Assumptions

I finally started talking to students instead of designing EBO entirely from my own assumptions. Two conversations were nowhere near validation, but they were enough to remind me why building before listening can be dangerous.

Read full article
EBO•Sep 25, 2026•18 min read

I Killed EBO

EBO was supposed to make studying easier to begin, sustain and return to. The smarter I tried to make it, the more tracking, context and attention it demanded—until the product started contradicting the reason I built it.

Read full article

Essays