Building an AI chatbot and the journey to deployment on Cloudflare
Originally published: 28 December 2023
This one is still eluding me, it's almost like coding can be hard sometimes
Introduction
In the ever-evolving world of AI and web development, I’ve embarked on an ambitious project: creating an AI Chatbot, a Chat GPT clone, utilizing the OpenAI API. The concept is straightforward yet innovative - a website that extracts video transcripts from YouTube, pushes them to OpenAI, and transforms them into comprehensive blog posts. Fascinating, right? However, this journey hasn’t been a walk in the park.
The Challenges: Streaming Responses and Lengthy Blog Posts
I encountered two major hurdles during the development process:
-
Lack of Streaming Responses: Initially, I found myself at a crossroads with no streaming responses. This obstacle proved more daunting than anticipated, as the available documentation offered little clarity or guidance.
-
Generating Longer Blog Posts: The second challenge was to coax OpenAI into delivering more extended blog posts, a feature crucial to the essence of my project.
The Search for Solutions
Determined to overcome these challenges, I plunged into an exhaustive search for answers. This phase was marked by trials and a steep learning curve, particularly with the integration of Firebase for storage, which turned out to be quite intricate.
A Breakthrough with Vercel
Then came a pivotal moment in my quest - discovering a Vercel template for a Chat GPT clone. It was a eureka moment! However, this template was primarily designed for deployment on Vercel and Netlify, with Cloudflare deployment presenting a unique set of complexities.
The Cloudflare Conundrum
The primary challenge with Cloudflare was its requirement for Server-Sent Events (SSE), necessitating the EDGE runtime. Implementing this on Cloudflare was anything but straightforward.
Triumph Over Trials
After days of relentless effort and tinkering, I achieved what seemed impossible - deploying the AI Chatbot on Cloudflare. The process involved:
- Adding
export const runtime = 'edge';to Multiple Pages: Contrary to the limited guidance provided, I had to incorporate this code into far more pages than initially anticipated. - Converting Dynamic Routes into Static Routes: This was a complex task that demanded patience and precision.
The Outcome
At present I have it deployed to cloudflare, but I’m seeing compatability issues… I’ll get there in time.
Why Does This Matter?
You might wonder why this endeavor matters. The answer lies in the broader context of my goals. Vercel, while an excellent platform, imposes charges for pro hosting, essential for monetizing any product. My ultimate aim is to reverse-engineer this Chatbot template and integrate it into my AI tools SaaS - another exciting project.
Deploying on Cloudflare not only saves costs but also opens the door to using Argo routing, which can significantly enhance Time to First Byte (TTFB). For more insights into hosting speeds, check out my comparison at sureshkhirwadkar.dev/posts/comparing-hosting-speeds.
In conclusion, this journey has been a blend of challenges, learning, and eventual triumph. It underscores the importance of perseverance in the tech world and opens up new avenues for innovation in AI and web development.
Read Count: 0
Facebook
Twitter
Email