Today is a milestone I’ve been working towards for a long time. I’m incredibly proud to announce that my first research paper, “A Serverless Architecture for Real-Time Stock Analysis using Large Language Models,” is now officially published and available on the open-access preprint server, arXiv.
You can read the full paper here: https://arxiv.org/abs/2507.09583
But this post isn’t just about the paper itself. It’s about the process—a process that I believe sits at the forefront of a new era in software development and research.
The Real Story: Using AI as an Execution Partner
From the beginning, this project was an experiment. The goal wasn’t just to build an AI stock analyzer, but to explore a modern workflow: could I, as an individual researcher, use my experience to guide a Large Language Model (LLM) in building a complex, serverless application from scratch?
I’ve always believed that the best tools are the ones that help us think and execute more effectively. For this project, I used Google’s Gemini not as a replacement for my own ideas, but as a powerful tool to bring them to life at an incredible speed.
My role in this collaboration was that of the architect and strategist. I designed the system, defined the logic, specified the features, and—most importantly—diagnosed the errors. I was the human guide, providing the “why” and the “what.”
The AI’s role was that of the tireless executor. It handled the “how”—generating the Python code, writing the HTML and CSS, drafting the LaTeX for the paper, and implementing the fixes I specified. This partnership freed me from the tedious aspects of line-by-line coding and allowed me to focus entirely on the bigger picture and the hard-to-solve problems.
The Project: A Zero-Cost, AI-Powered Dashboard
The result of this human-AI partnership is a fully functional system with two key components:
-
An Automated Backend: A Python script, running on a free schedule via GitHub Actions, fetches daily stock data, gets qualitative analysis from the Gemini API, and calculates performance metrics.
-
A Live Frontend: A clean, responsive dashboard that visualizes the AI-generated insights, including sentiment, a predicted price range, and a daily accuracy check.
The entire system operates at a near-zero cost, making it a viable blueprint for students, hobbyists, and independent developers everywhere.
-
See the Live Dashboard: https://codepen.io/tanivashraf/pen/GgpgxBY
-
Explore the Source Code: https://github.com/TanivAshraf/ai-stock-analyzer
The Unseen Challenge: A Case Study in Real-World Debugging
A major part of my research paper is dedicated to the debugging journey. We often see polished final products, but rarely the struggles required to get there. My paper documents every critical bug we encountered, from simple Python TypeErrors to a bizarre, platform-level GitHub bug that required rebuilding the entire repository from scratch to solve.
This transparent account of the development process is, I believe, one of an early and few examples of a published case study detailing this new, collaborative way of building and problem-solving with AI.
My Lesson: Using AI in a Humble and Powerful Way
This project solidified a core belief for me: AI’s greatest immediate potential isn’t in replacing human ingenuity, but in amplifying it. By using these tools in a humble way—as partners to execute our vision—we can achieve incredible things. It’s about being the driver, with AI as the engine.
I hope this project and the accompanying paper provide a practical and inspiring blueprint for others. I’m already planning my next project, where I intend to automate even more of this process. The future of human-AI collaboration is bright, and I’m excited to be a part of exploring it.