Ranga Reddy Nukala

Hey! I'm Ranga Reddy Nukala. I'm in the Applied Data Intelligence MS program at SJSU, focused on LLMs, physical AI, and robotics. Before grad school, I was a Senior Software Engineer building iOS apps with Swift and SwiftUI - two-time Apple Swift Student Challenge winner (WWDC'20 & '21).

This blog is where I write up experiments, reading notes, and things I built. Topics are mostly ML/AI with occasional Swift posts.

Posts

How voice agents decide when to listen, think, and speak

Voice agents have to make one decision over and over again: should I be listening, thinking, or speaking? It sounds trivial. Humans do it without noticing. But in real-time audio, it’s where most “the bot keeps cutting me off” and “the bot just sits there” failures come from.

Why My AI Agents Were Stuck

I run a multi-agent system with 10 AI coding agents. A project manager, four platform-specific developers, and several specialists. They pick issues from a GitHub Project board, implement features, open PRs, and report status through Discord. Cron jobs fire every two hours. Heartbeats pulse every thirty minutes.

Building Wishper: Local Voice-to-Text for macOS on MLX

I built Wishper, a local voice-to-text app for macOS that runs entirely on Apple Silicon. Hold a key, speak, release — cleaned text appears in your active app. No cloud. No subscription. 90MB app bundle.

Debugging a Voice AI Receptionist: Five Calls, Four Failures

I set up an ElevenLabs Conversational AI agent to autonomously answer phone calls: greet the caller, figure out why they’re calling, collect details, confirm, and hang up. The first real call was a disaster — the agent repeated its greeting three times until the caller hung up. Debugging the conversation transcript revealed two silent config overrides from the platform.

3D Pose Detection for Plank Form Analysis on iOS

I built a plank workout tracker that uses the front camera to detect what type of plank you’re doing, measure your body angle, and flag when your hips sag or pike — all in real-time. The core of it is Apple’s VNDetectHumanBodyPose3DRequest, which gives you 3D joint positions in meters. The interesting part was turning those positions into useful form analysis.

160 Commits in 24 Hours: Refactoring a Production iOS App with Xcode's Agentic AI

DEW (Daily Eastern Wisdom) had been sitting untouched for months. It worked, but the internals were deteriorating: a 1200-line DataManager handling everything, Core Data fetches on the main thread, competing cache systems, and near-zero test coverage. When Xcode 26.3 shipped with agentic AI via MCP, I opened the project to see how far the tooling could push a real refactor. 160+ commits later, across ~24 hours of development time, the app shipped as v0.6.2.

Cleaning Up Diamond Dependencies in Swift Packages

My app’s modular architecture had a dependency mess: Account’s Package.swift declared 5 explicit dependencies, but 4 of them were already pulled in transitively through Events. I cleaned this up across several packages by trusting SPM’s transitive dependency resolution.

CoT vs CoD: When Does Multi-Agent Debate Actually Help?

CoT vs CoD Banner

Physical AI: What I'm Paying Attention To

NVIDIA GR00T N1