Rebuilding My Failed Startup Idea: Introduction
In June 2023, my 2 cofounders and I set out on a journey across the United States to bring our startup idea to life. Fresh out of college, we were accepted into an accelerator program and decided to move to New York for 4 months to pursue our dream.
Our startup, Bench AI, aimed to be a closed-source platform that simplified training ML models on the cloud without the complexities of configuration. The concept was exciting, and working on Bench AI was truly enjoyable. However, we faced a significant challenge - we had no users. We had fallen into the common startup pitfall of overengineering a product without a market.
The Journey of Bench AI
Initially, we targeted large corporations, claiming our tool could revolutionize their DevOps processes. Unfortunately, our pitch didn't resonate with them as they preferred to rely on their in-house teams. We then shifted our focus to startups, but encountered similar challenges. The few companies training their AI models had well-established infrastructures and were not interested in our solution.
After multiple pivots, we finally aimed our product at solo ML engineers, researchers, and developers. However, this audience had limited resources and preferred more established solutions. Despite our efforts, we were unable to gain traction.
A Turning Point
Feeling lost and seeking guidance, I reached out to a mentor in the developer tools industry. He advised that charging for our service was hindering our growth and suggested adopting an open-source approach. According to him, open-source adoption would validate if our product was addressing a real need. Inspired by his insights, we decided to pivot once again.
Fast forward a year, with a new team and a fresh idea, we transitioned to the realm of digital accessibility. Our current focus is on developing tools that help users ensure their documents comply with accessibility laws using AI technology.
Reflecting on the Past
While I am grateful for the progress we've made in our new venture, I can't help but wonder what could have been if we persisted with Bench AI. As a developer, creating dev tools brings me immense joy. Recognizing the continuing need for automated cloud infrastructure, especially with the rise of open-source AI models, I've decided to revisit the original Bench AI concept in my spare time.
The Future of Bench AI
Learning from past mistakes, I plan to redevelop Bench AI, emphasizing open-source principles. The new version will focus on inference rather than training, as it aligns more with market demands. Additionally, I will implement a serverless architecture to optimize costs and scalability.
The revised Bench AI will feature a simple yet scalable architecture, involving components like a reverse proxy, an HTTP server, and a message queue. I aim to build a network that is self-hostable and easily accessible via a REST API.
Looking Ahead
While there is much work ahead, I am excited about the journey. Stay tuned for future articles detailing the development of Bench AI, including a deep dive into the message queue implementation.
