1. Building a Multi-Modal Chat Application

We envisioned a chat assistant that could guide learners through complex projects—answering questions, reviewing code, checking diagrams, or helping troubleshoot hardware setups. That meant building a multi-modal AI chatbot that could understand and respond to:

Text (questions, instructions)
Images (circuit diagrams, handwritten notes)
Context from the user's ongoing project

Pulling this off was no small feat. We had to:

• Maintain a rich context memory per user, so the bot could give meaningful help, not generic answers.
• Balance between being technically accurate and conversationally engaging—especially in educational or corporate settings -where tone matters.
• Making the chat feel like a true mentor, not just a smart FAQ bot, took time and tons of user testing.

We managed to implement it successfully with 2 AI based chatting, one a simple chatting version and another multi model application.

2. AR Interaction on Smartphones

Creating meaningful AR interactions on smartphones—without relying on high-end AR headsets—was one of the toughest nuts to crack. We wanted users to be able to build and interact with their projects using just what they had: a phone, some basic hardware, and maybe even scrap materials like newspapers or ESP32 boards.

But AR on smartphones comes with serious limitations:

• Tracking physical objects accurately in real time, especially in low-light or cluttered environments, was unreliable.
• Performance dropped sharply on budget devices, making experiences laggy or outright unusable.
• We had to fine-tune our AR layers, simplify interactions, and use clever tricks like marker-based tracking and lightweight 3D rendering to deliver a smooth experience across most devices.
• In the end we were not able to implment this properly, the AR detection and drawing is extremely slow