Most of these are unfinished. Each one helped me figure out what I wanted to work on next.
ps one must drop a project if they are not obsessed about it
Simulate a world of characters with their own goals and memories. Let their choices create the story.
Simulate a world first; extract the story afterwards. Characters with their own goals, beliefs, and memories, whose choices create the plot.
Unfinished prototype exploration. Consistently good long-form stories remained out of reach.
Bring fragmented geopolitical signals together to estimate the risk of escalation before it becomes obvious.
Could many weak signals reveal geopolitical escalation early? I built a pipeline that brought conflict, economic, diplomatic, and market data into a probabilistic risk model.
Substantial prototype, paused. Forecast accuracy and the later trading idea were never validated.
Trace a wrong maths answer through its prerequisites. Diagnose the root skill gap before deciding what to teach.
Why did the student get it wrong? A maths diagnostic that traced mistakes through prerequisite skills to find the root gap and suggest what to learn next.
Diagnostic design and prototype exploration. The inferred gaps were not validated with enough real students.
A website with one big button that plays the FAHHH meme sound. Sometimes the thing to make is just a button.
Try the buttonAn assistant for dinner, travel, appointments, and everyday routines. You provide intent and judgment; it handles the coordination.
An assistant that takes intent and handles the coordination: dinner, travel, appointments, and everyday routines. I explored an agent runtime, a Watch interface, and an AI + human concierge.
Product and architecture exploration. It became a possible interface over a deeper personalization system.
Turn raw signals into meaningful current state. A real iPhone location test successfully recognized that I was at the office.
For Alfred to help proactively, it needed to know what was happening. This layer turned raw signals into current state; a real iPhone location test successfully recognized that I was at the office.
Working technical prototype, unfinished. Knowing where I was still didn’t tell an agent what it should do.
Preserve product intent across long building sessions: a conversation becomes a PRD, approved tasks, agent runs, and human review.
A place to preserve product intent while coding agents execute: conversation to PRD, approved tasks, agent runs, and human review. Product truth stayed separate from code state.
Local prototype, paused. Review handoffs, stale state, and recovery needed more work.
Capture a screenshot or a piece of text, add a small thought, and find it later by asking your memory.
Capture a screenshot or a piece of text, add a small thought, and find it later by asking your memory. A Mac app that kept the original evidence alongside AI enrichment.
Usable Mac prototype / early beta, later inactive. Reliable resurfacing and cloud/mobile sync remained unfinished.
An animated mini-series about Kunal, 26, in Bangalore: almost disciplined, almost over her, almost sorted. Funny first, then a small emotional bruise.
“Life is happening. He’s still waiting.” An AI-assisted animated mini-series about Kunal, 26, in Bangalore: almost disciplined, almost over her, almost sorted. Funny first, then a small emotional bruise.
Character, visual direction, and episode concepts developed. It didn’t become a sustained published season.
Can a persistent understanding of a person improve decisions beyond current context, relevant memories, and asking one good question?
Can a persistent understanding of a person improve decisions beyond current context, relevant memories, and asking a good question? The question that grew out of Alfred and the Context Inference Engine.
Active research thesis, unresolved. Relevant episodic retrieval proved strong; several dataset routes closed without establishing that persistent modelling beats it.
A desk that felt too consequential,
but never turned out to be.
Long, repetitive days in coaching classrooms. A lot of pressure, and very little sense of who I was outside the goal. I didn’t get into IIT.
Learning to admit when something isn’t working started here.
I arrived feeling behind.
Then I found the workshop.
Robowars. Welding-shop runs. Fried components and midnight searches for parts. In my second semester, I found my way to an internship at Accio Robotics.
One night before a pilot, I broke the 3D-printed parts. I thought I’d get fired. What mattered was figuring out how to fix them.
A laptop and a chair.
My version of “a chip and a chair.”
Blender, CAD, hackathons, crypto, trading, podcasts. I even started my own philosophy podcast. It lasted one episode. I was looking for opportunities where a little effort could go a long way.
Then things got quieter. I slowed down, finished college, and let there be a pause.
Making the machine was
only part of making it work.
I joined Accio full-time and moved between assembly, design, planning, and deployments. Then came a product launch, an expo, rebranding, and operations.
How do you ship an unfinished product across the country in pieces and get them all to the expo at the same time?
We couldn’t keep exploring
every direction.
Funding conditions forced a narrower focus. I worked across solution design, pre-sales, and product research, searching for a problem the company could meaningfully solve.
The bet became dark-store automation.
A gantry system.
And everything it took to get it built.
Development, manufacturing, supply chains, R&D, deadlines, and whatever broke that day. I drove the dark-store automation product end-to-end.
By January 2026, customer and acquisition interest was emerging. I chose that transition as my moment to leave and figure out my own next move.
The present · 2026
First, I slowed down. Coffee, cooking, chess, poker. Messing around with AI without needing it to become something.
Gradually, the curiosity became a search again.
Projects explored in 2026Take the big swing now, or build toward it?
Back to the present
