Shard AI

Inspiration

Shard AI was inspired by Vannevar Bush’s Memex, a vision of a machine from the 1940s that could store and link all knowledge. I’ve always been obsessed with the idea of extending human memory. I wanted to create something that could remember not just important thoughts like business ideas but also the random, off-the-wall moments. Whether it’s deep research or something as random as, “Holy shit, I realized I like thick latinas more than white women,” I wanted an AI that could hold it all.

What it does

Right now, Shard acts as a hybrid between a note taker and a Perplexity for notes. It stores everything, whether it’s your deep research, random ideas, or personal insights. But the real magic happens when it starts to talk like you. When you ask Shard about something you’ve stored, it doesn’t just spit out a static note. It processes everything in context, responding like it’s you talking. This gives it the feeling of a digital clone, one that knows you, your thoughts, and how you think.

It uses embeddings to make sure your notes are connected and easy to retrieve, and RAG ensures that it pulls the most relevant information to generate responses that feel both natural and personalized. It’s not just a tool for remembering; it’s a way to bring your thoughts back to life with a little extra help.

How we built it

I’m not a coder. I had zero experience with programming before starting this project—didn't even know how to write "Hello World" in Python. But I wanted to push myself and learn something new, so I dove in and started building Shard with Bolt. I used Supabase for the database, storing all my notes and thoughts securely, and hosted the backend on Vercel. The whole setup was pretty much a learning curve, and I spent a lot of time figuring out how to get everything connected and running smoothly. For the tech side, I mainly used JavaScript and TypeScript, which were both new to me. The real breakthrough came when I started exploring embeddings to connect my thoughts and RAG to help Shard pull the right notes and create responses that felt like me. This was the tricky part—getting Shard to not just recall a note but also to generate answers in a way that seemed natural and context-aware. I also integrated the OpenAI API to power the conversational AI, making sure it could generate realistic, human-like responses. For the UI, I took inspiration from Apple’s liquid glass aesthetic.

Challenges we ran into

The biggest challenge was definitely starting from scratch with no coding experience. Debugging was a constant struggle, especially when things didn’t go according to plan. I spent countless hours figuring out how to make the embeddings work without slowing the app down, and ensuring the RAG system pulled relevant notes quickly and efficiently was tricky. Integrating everything was also a huge challenge. Getting the frontend and backend to communicate seamlessly while making sure the AI responses felt natural and personal required a lot of trial and error. There were also issues with the API calls, as things didn’t always sync up the way I expected.

Accomplishments that we're proud of

One of the things I’m most proud of is that I built Shard AI from scratch, despite having zero prior coding experience. I never thought I could be a builder, let alone an entrepreneur. I never imagined I’d be able to create something people actually want, like YC would say. But here I am, using my own app and seeing two of my friends really enjoy it, find it helpful, and use it daily and I’ve only told like five people about it. Getting the RAG system to work seamlessly and making the AI respond like a digital clone of me was another big win. The fact that my own creation is being used and appreciated, even in its early stages, really pushes me to keep going.

What we learned

I now actually believe Sam Altman’s quote: “You can just do things.” I had zero coding experience when I started this project, and the idea of building something technical felt impossible. But I proved to myself that it’s possible to dive into something new, learn, and make it work. Now, I’m just pumped to keep building shit I find cool. I’m sure I’ll win no matter what, even if it takes building 1,000 more projects. It’s all about creating and learning along the way, no matter how many failures it takes.

What's next for Shard AI

Next up, I’m focused on building out the memory feature properly, so Shard can store and recall even more meaningful data, making it feel more like a true digital clone. I also want to add voice notes and conversational AI using the Eleven Labs API, so users can have a more natural, interactive experience with Shard. I’m planning to launch the MVP to users as soon as possible to get real feedback and iterate quickly. I know the idea behind Shard isn’t exactly groundbreaking, and it’s a pretty common concept, but I’m ready to give it my all and try my best to make it stand out.

Built With

• bolt
• claude
• javascript
• openai
• supabase
• typescript
• vercel