Writing
Blog
6 posts on AI agents, voice pipelines & edge systems.
I Tried Running a Local LLM on My Android. Here's What I Found.
Llama.cpp was unusable. Google AI Edge Gallery was surprisingly good. And now I know exactly why — and what to build next.
Your Prompts Are Terrible Because You're Typing Them
Most people blame the model. The problem is always the prompt. Here's how to fix yours.
Kitchen Surveillance with OpenClaw: Person Detection Under 50MB RAM
How I built a real-time kitchen surveillance agent on a Redmi Note 7 Pro using MobileNet-SSD, OpenCV, and TFLite — detecting people under 50MB RAM and sending Telegram photo alerts.
I Installed OpenClaw on My 6-Year-Old Redmi Phone. It's the Best AI Agent Experience I've Had.
How I turned a drawer-forgotten Redmi Note 7 Pro with a cracked screen into a 24/7 AI agent using OpenClaw, Termux, LineageOS, SSH, PM2, and DeepSeek — and why it beat running agents on servers.
Building an AI Agent Orchestration System That Shipped to 200 Users on Day One
How I architected Champion's five-agent investor simulation platform — multi-agent design patterns, second-brain memory systems, AWS infrastructure, and shipping to 200 users on launch day.
Real-Time Voice AI: Building Speech-to-Speech Agents with Sub-Second Latency
How I reduced Claritel's voice agent initial response time from 8s to 3s — covering WebSocket Architecture, Pipecat, Deepgram, ElevenLabs, and the specific optimizations that cut latency by 62%.
Written by Mohd Mursaleen — AI agent engineer & full-stack developer, Bengaluru.