PDFラク
A multilingual PDF utility site for merging, compressing, converting, editing, OCR, and benchmarked browser-side document workflows.
Browser-first PDF toolsNauman Mustafa
AI systems engineer, former senior machine learning engineer, and indie builder.
I work on OCR, computer vision, LLM workflows, and software delivery with coding agents. This site is a small notebook for projects, field notes, and technical essays from that work.
Projects
Offline speech to text
A private, multilingual live-transcription project that runs Whisper locally in the browser without uploading microphone audio.
Browser WASM experimentTranslateX
An iOS-first real-time voice translation app for English and Japanese conversations.
Shipped personal appWriting
Building a Personal AI Workflow: From Reading to Private Video
A field report on building a human-supervised AI workflow that turns starred inbox reading into natural narration, visual slides, and private video.
Why I'm Sticking With GPT-5.5 Over Claude
A personal comparison of GPT-5.5, Claude, Codex, and Claude Code based on coding workflow, context handling, readability, limits, and developer experience.
Running a YouTube Channel on Autopilot
A 2-month YouTube automation experiment covering news scraping, video generation, auto-uploading, niche choice, costs, and watch time.
Gaining 4 Years of Software Engineering Experience in 4 Months
AI changed my job from writing every line to directing parallel software systems. This is my field report from that shift.
From Vibe Coding to Compound Engineering
A software engineer journey from vibe coding to compound engineering, and what AI changes about architecture, intent, and code quality.
Lessons Learned Vibe Coding an iOS App from Scratch
Field notes from building an iOS app with AI coding tools, covering what worked, what failed, and how to ship faster with quality.
MLUI Mobile: Autify OCR vs. Google OCR
A benchmark of Autify OCR versus Google OCR on rendered mobile UI text, including methodology, accuracy metrics, and failure analysis.
Token Compression: Reducing Attention Waste?
An experiment on compressing multiple LLM tokens into one representation for faster decoding and longer effective context.
A quest for very long context: Part 1
Experiments on extending transformer context length, including training observations, tradeoffs, and lessons from long-context tuning.
Machine Learning Features in Autify for Mobile
How Autify for Mobile uses machine learning for visual regression testing, visual self-healing, and AI-powered app exploration.