2024 Wrapped: Building 7 AI Projects – From Document Search to Immigration Tools
As 2024 comes to an end, I’m taking a moment to reflect on my year of building seven AI projects—all while working a full-time job. This isn’t one of those stories about hitting big milestones like $1M ARR or going viral on Product Hunt. It’s more about what happens when you follow your curiosity, work on ideas that excite you, and enjoy the process itself. From tackling server timeout issues in my first RAG (Retrieval-Augmented Generation) project to building a Tesla referral tool over a weekend, and finally making something I really needed for my own immigration journey—it’s been a year of growth and learning.
If you’re a developer, an indie hacker, or just curious about how to build multiple projects in a year, I’ll share the lessons I learned, the tech decisions I made, and why scratching your own itch can lead to the most fulfilling products.
Diving into RAG: Building searchmydocs.ai
In October 2023, I became fascinated with RAG (Retrieval-Augmented Generation). While many were already busy creating ChatGPT wrappers, I knew I was late to the game—but I wanted to dive in and learn anyway.
Over Thanksgiving, I stumbled upon an incredible two-hour tutorial by Greg (@ggrdson) from Supabase that explained how to build RAG systems step by step. Inspired and excited, I decided to build searchmydocs.ai to put those concepts into practice.
Here’s the catch—I wasn’t very confident in frontend development. I spent a week learning React.js and relied heavily on ChatGPT to help me bridge the gaps. I even tried convincing some frontend friends to partner with me, but no one shared the same excitement for building an indie SaaS. Looking back, it was a blessing in disguise—it forced me to figure things out on my own and grow in the process.
The journey wasn’t without hurdles. At first, the system would fail when processing PDFs larger than a few pages, hitting CPU timeouts during embedding generation. Greg’s tutorial was fantastic, and I initially wondered why he was giving away so much knowledge for free. But once I started implementing it, I realized there were limitations—particularly with serverless workers struggling under heavier loads.
To address these issues, I switched the embedding generation to OpenAI’s API instead of native serverless workers. This improved the functionality somewhat, but the overall engine still wasn’t as robust as I wanted.
Even with its imperfections, this project was a huge learning experience. I set up an LLC, joined Microsoft for Startups, and even got my first payment from a stranger. Building a RAG engine as a first SaaS felt like running a marathon while still learning to walk, but it was an invaluable journey that taught me so much.
https://x.com/sreeenidhi/status/1743740144912584810
https://x.com/sreeenidhi/status/1752733697781203306
https://x.com/sreeenidhi/status/1755416671781781767
https://x.com/sreeenidhi/status/1758698057552785430
Exploring AI Images: roomreimagined.com and imageenhancerai.com
After the highs and lows of working on RAG, I wanted to try something fresh and park searchmydocs for a bit. AI image generation caught my attention, and seeing interiorai.com by @levelsio thriving made it clear—this idea had already been validated. It was the perfect opportunity to dive into the world of image-generation products, so I decided to take a shot at creating an AI-powered interior design tool.
I had purchased the domain roomreimagined.com back in October 2023 but decided to park it until I finished shipping searchmydocs.ai. Once that MVP was out the door, I was excited to dive into the next project and finally bring roomreimagined.com to life.
The timing felt perfect. The best part? I didn’t have to start from scratch. Someone on Replicate had already developed the core technology I needed, so I could focus on simplifying the user experience and making it accessible. I also decided to skip the usual subscription model and went with a pay-as-you-go credit system, letting users pay only for what they needed. This approach kept things flexible and user-friendly.
By the time I launched, the market was crowded with similar tools. That didn’t bother me. Over time, I’ve learned that if I’m genuinely excited about building something, it’s worth doing regardless of how saturated the space is. That excitement and momentum carried over to imageenhancerai.com, a broader tool for enhancing AI-generated images.
Each project built on the experience of the last, helping me ship faster. I was enjoying the process and getting better with each step.
Making Tesla Referrals Easier: teslareferralhub.com
Next up was teslareferralhub.com. Tesla’s referral program is amazing, but there’s a catch—you only really earn rewards if you have a strong social media following. I wanted to create something where anyone could participate, even if they didn’t have a big online presence.
So, I built a platform where users could submit their referral links for $20. These links were added to a queue and displayed in a round-robin format, ensuring everyone got a fair shot. By this time, Claude had become incredibly good, making it super fast to put together a UI. I managed to build this project in just a weekend.
Shaan Puri recently mentioned something on the MFM podcast about Mike Posner that really resonated with me. After years of chasing his past successes, Mike realized the key is to do what you think is cool. Sometimes, a large portion of the population agrees, and it takes off in ways you couldn’t predict. For me, this project felt exactly like that—I thought building a referral-sharing platform was cool, so I went for it. I had a blast putting it together, and the process itself was incredibly rewarding.
Expanding the Concept: referralsearchengine.com
After teslareferralhub.com, I wanted to take the concept to the next level—a generic platform for sharing referral links, not just for Tesla, but for any product. In the SaaS world, it’s common advice to start with a niche, focus on making it work, and then expand. Think about how Amazon started with books before becoming the everything store. Even though teslareferralhub.com wasn’t a massive success (though I did get two referral bonuses!), I believed in the idea enough to build a broader version of it.
That’s how referralsearchengine.com was born. This time, I built it on Cloudflare’s stack, mostly because I liked their free tier better. The platform allowed anyone to share one referral link for free, no matter the product. If they needed more links, they could buy additional slots.
This project gave me some great experience with Cloudflare Workers and made me appreciate their power and flexibility. By this point, I’d become more efficient with shipping products, and each project was helping me get faster at putting ideas into action.
However, one key thing I realized with teslareferralhub and referralsearchengine is that these kinds of platforms live and die by distribution. And here’s the thing—I don’t find joy in the grind of improving distribution through SEO, building backlinks, or manually promoting links across different platforms. I love building products, but the marketing side just doesn’t excite me as much.
Returning to Document Analysis: secanalytica.com
My next product was secanalytica.com, a tool that leverages OpenAI’s Assistants framework. By this point, the costs of using the Assistants API had come down to manageable levels, making it feasible to build a scalable product. Most of my time on secanalytica was spent creating a robust wrapper around the framework—something versatile enough to be applied across different domains.
For the initial use case, I chose SEC filings. The idea came after seeing tweets from @virattt, who was working on financial datasets. The domain seemed rich with possibilities, and I knew there was enough value to justify focusing on this area.
With secanalytica, users can upload SEC files, which are then analyzed to provide actionable insights. The wrapper I built has the potential to expand into other domains in the future, but for now, SEC filings serve as a great starting point.
The next logical step is to implement report generation using RAG (Retrieval-Augmented Generation). However, I’ve temporarily paused this feature because another project caught my attention and felt too exciting to delay. I plan to circle back to secanalytica soon and pick up where I left off.
Building for a Personal Need: visamonkey.com
The next project on my journey was visamonkey.com, a document management tool designed for immigrants and non-immigrants alike. It’s a platform for managing visas, passports, H1B documents, and more, with features like intelligent search, timeline tracking, deadline reminders, chat with documents, and application tracking.
This project is still a work in progress, but it holds a special place in my journey. I built it because I needed a solution for myself and for others who face the same struggles. As someone who has dealt with the challenges of visa documentation, I know how overwhelming it can be to stay on top of deadlines, keep everything organized, and avoid mistakes that could have serious consequences.
Visamonkey was born out of that experience. Building something to address my own problems brought a lot of clarity to the process—every feature felt obvious because I understood the pain points deeply. While there’s still a lot to do, I’m excited about the potential of this platform to make life easier for people navigating the complexities of immigration and document management.
Launching the Future: raglauncher.com and thirdbrain.io
With the RAG wrapper now in place, my next project is raglauncher.com—a boilerplate for launching RAG applications quickly using OpenAI’s Assistants framework. The goal is to simplify the process of creating robust RAG-based tools, enabling developers to focus on building solutions rather than reinventing the wheel.
Raglauncher will serve as the backbone for multiple projects, including secanalytica.com, visamonkey.com, and thirdbrain.io. Each of these projects showcases the flexibility of the wrapper applied to different domains, from SEC filings to immigration management.
Reimagining SearchMyDocs as ThirdBrain
As part of this journey, searchmydocs.ai will be rebranded as thirdbrain.io. The vision for ThirdBrain is ambitious—it aims to evolve into a platform that learns across various dimensions of a user’s data and helps provide clarity of thought. While the specifics are still abstract, the goal is to create a tool that goes beyond traditional document search, offering deeper insights and understanding.
A Unified Vision
With raglauncher.com, the idea is to bring everything I’ve learned into a cohesive system. It’s the culmination of experiences from building specialized tools for different domains and figuring out what works best. Each project, whether secanalytica, visamonkey, or thirdbrain, will benefit from this shared foundation, ensuring faster iterations and consistent improvements.
I’m excited to see where this next step takes me and how it can empower others to build their own RAG-based applications efficiently.