From persistent memory for AI agents to ideas that become physical objects. I’m interested in what happens when curiosity becomes something you can use.
The repository contains the implementation, architecture, and documentation. The central question: how can an agent retain useful context across interactions? Memory quality and retrieval tradeoffs need evaluation; persistence alone isn’t proof of better recall.
PERSISTENT CONTEXT / CONCEPT
An interaction
↓
01Capture context
02Organize memory
03Retrieve what matters
↓
A more informed next step
CONTEXT THAT OUTLASTS THE CONVERSATION
PERSONAL AI / SECOND ITERATION
Annie in a Bottle. AI in everyday life.
Annie in a Bottle is the private second iteration of my personal AI assistant, following an open-source first version. Connected to Home Assistant, Annie helps with everyday routines, tracks my water intake, and provides proactive support throughout the day.
Why Annie matters to me +
This project brings AI into the rhythm of daily life: connected-home context, personal routines, and ongoing conversations. I use it to explore what an assistant should remember, when it should proactively help, and when it should ask. The current iteration is private; the earlier version is open source. Even this website began with a conversation with Annie.
DEVELOPER TOOLS / OPEN SOURCE
02↗
Better tools for the act of building.
A collection of specialized AI agent definitions spanning planning, architecture, implementation, review, and testing. An exploration of how AI can support the engineering workflow.
3D printing is a passion that grew from an Ender 3 to a Voron 2.4 I built from scratch—using the Ender to print parts for the Voron. Today I also build with Bambu Lab printers and design models that others can make their own.
Thousands of downloads and prints of my 3D models.
Building the Voron meant getting close to the machine itself. Designing and sharing models carries that curiosity forward: an idea becomes a physical object, and then something another maker can print. I’m also exploring AI-assisted modeling and printing, with human judgment and iteration throughout the process.
02 / HOW I THINK
Curiosity needs a workbench.
The idea is the beginning.
01
Make it useful.
Start with a problem worth solving. Build something people can actually put to work.
02
Stay close to the work.
Move between the architecture and the details. Prototype, inspect, question, and improve.
03
Let reality give feedback.
A working demo is a beginning. Reliability, limitations, and what happens in practice belong in the story too.
03 / THE PERSON BEHIND THE WORK
A builder’s curiosity. A leader’s perspective.
I’m Ankit Agarwal, a Director of Engineering at Intuit and a hands-on builder interested in AI, software systems, and physical making.
My professional work includes fintech platforms and AI-enabled products. My personal projects give me room to explore new tools, test ideas, and follow a question all the way to a working prototype.
This is a home for that work: what I’m building, what I’m learning, and the ideas that keep pulling me back to the workbench.