Mahdi
ML Engineer & Softwere developer
I build systems that think — from voice-based medical screening to multi-agent tooling.
About
I like problems that sit at the seam of machine learning and real products. I run my own infrastructure (a little Ubuntu box named rose), ship side projects end-to-end, and care about the craft as much as the result. This site is one of those side projects.
- ML engineer focused on applied deep learning and agentic systems.
- Self-hosts a home server ('rose') and runs parinox.ir.
- Flagship work: Parkinox (voice-based Parkinson's screening), Yolda (offline-first travel app), AgentWire (multi-agent framework).
Projects
Parkinox
ML/AI — Parkinson's disease detection from voice biomarkers. End-to-end pipeline: audio feature extraction, model training, and a clinician-facing dashboard for screening.
Stack: Python, PyTorch, FastAPI, React
SourceYolda
Mobile — Companion app for road trips — offline-first routing, points of interest, and a shared trip log. Built for spotty-connectivity travel.
Stack: React Native, Expo, SQLite, TypeScript
SourceAgentWire
ML/AI — Framework for wiring multi-agent LLM workflows with typed tool contracts, retries, and observability baked in.
Stack: TypeScript, Node, LLM, Zod
SourceRose OS
Open Source — This site — a personal portfolio built as a working operating system. Desktop window manager, mobile OS shell, live GitHub stats.
Stack: Next.js, Zustand, Framer Motion, Tailwind
SourceLiverose homelab
Backend — Self-hosted Ubuntu server ("rose") running personal services behind a reverse proxy, with automated backups and monitoring.
Stack: Ubuntu, Docker, Nginx, Prometheus
SourceTerminal
Archived — A terminal app that displays the bio of the user.
Stack: Next.js, Zustand, Framer Motion, Tailwind
Source