- Atul for Marketing
- Topics
- Artificial Intelligence
Artificial Intelligence
Projects powering or powered by AI

Building a RAG-Powered AI ChatBot from Scratch
This Project52 build was a deep dive into creating a fully functional, multi-tenant AI chatbot platform using Retrieval-Augmented Generation (RAG). The goal: allow any business to train a chatbot on their internal documents, deploy it publicly, and capture leads all while keeping their data completely isolated and secure. We built everything from scratch including user authentication, lead classification, file-based knowledge ingestion, and admin dashboards combining FastAPI, LangChain, SQLite/PostgreSQL, and a custom React frontend. This wasn’t just a chatbot; it was the foundation for a scalable SaaS product that gives each business its own smart assistant.


DQN vs PPO: Which Reinforcement Learning Algorithm Lands Better?
Using OpenAI Gym’s LunarLander-v3 environment, I trained two AI agents from scratch: one using Deep Q-Networks (DQN), a value-based method that learns which actions lead to maximum reward, and the other using Proximal Policy Optimization (PPO), a policy-gradient method that directly learns optimal behaviors. I tracked their learning curves, reward scores, fuel usage, and even visual behaviors to determine which agent could not only land but do so gracefully, efficiently, and reliably.
