Labs

Things I've built, tested, and obsessed over.

AI Nature Camera

Not Open Source

From 40 Seconds to 3 Seconds, Performance Engineering on Mobile

40s Before

3 - 5s After

~90% Faster

An AI-based Nature Camera app that combines real-time preview, ML-based subject segmentation, and a set of portrait-style effects Invert, Swirl, Zoom Burst, Duotone, Phantom, and more. It also uses background processing so photos continue saving even if the app is closed. The biggest lesson here wasn't about features, it was about understanding exactly where every millisecond and megabyte was going.

Engineering Optimizations

Resolution-Aware Processing - Matched the ML and effect layers to the necessary output scale.
Memory Management - Drastically reduced unnecessary memory copies between the JVM and native layers.
Math Optimization - Refined blur and pixel operations to run more efficiently across the processing pipeline.

Flutter Kotlin Native Android CameraX OpenGL ML Kit TensorFlow Lite Computer Vision Vibe Coding

Aria

Live Project

Multi-Agent AI System with Intent Routing

4 Agents

1 System

Real-Time Routing

Aria is a multi-agent AI system that routes user input to different agents based on intent. Instead of a single chatbot, the system decides how to respond using structured routing logic and fallback AI handling. It is designed to handle real-world inputs including mixed queries, conversational context, and priority-based decisions.

System Design

Priority Routing - Security-related inputs are always handled first when multiple intents are present.
Hybrid Intent Detection - Combines rule-based logic, conversation detection, and AI fallback.
Multi-Agent Architecture - Separate agents for security, recommendations, behavior analysis, and chat.

Python FastAPI Async IO LLM Integration NLP Intent Routing

Try the live system →

Something New is Brewing

The next project is currently in active development. It's going to be interesting, check back soon.

In Progress