I build agents that cite their sources, ask before they act, and survive a restart.
I'm Saleh (Mohammad Saleh Mahdinejad on paper), a full-stack developer and computer engineering student at the University of Isfahan. Most of what I build has a model somewhere inside it, but the part that decides whether it holds up is everything around the model: retrieval that finds the right page, a queue that survives a crash, a confirmation step before anything irreversible.
I work in Persian and English, and so do the tools I make.
Your coding agent can now direct films. An agent skill for Codex, Claude Code and friends: a motion-graphics kit that cuts a film into one-idea shots, 256 techniques, 32 transitions, a from-scratch 3D renderer and a sound engine. Every frame is a pure function of time, so an agent can render one, look at it, fix a number and try again. Benchmarked against itself, with and without the skill.
|
Hear every voice, or just read it, in your language. Live translation and dubbing for desktop and browser audio. A Chrome and Edge extension and a desktop app for Windows, macOS and Linux, with original audio, dub, source subtitles and translated subtitles as four independent outputs, in more than thirty target languages. No telemetry.
|
Multilingual research, grounded in evidence. A self-hosted workspace that turns documents, media, notes and web pages into cited conversations, slides, audio and video. Agentic RAG that says so when the sources don't support an answer. English and Persian, left-to-right and right-to-left.
|
A news desk that runs on your own machine. Local-first news collection and reporting for subjects you define. Websites, GitHub, Reddit, Telegram and X in, a bounded multi-provider LLM router in the middle, grounded reports out on Telegram, in English or Persian.
|
|
A Persian insurance comparison and decision platform. Quotes from several sources side by side, with recommendations that explain themselves. An assistant that works through controlled tools, by text or by voice. Installment analysis.
|
| Languages | Python TypeScript JavaScript C# C++ Java |
|---|---|
| Back end | FastAPI Django DRF Channels Node.js Express .NET Celery BullMQ |
| Data | PostgreSQL Redis SQLite FTS5 Chroma |
| AI | LLM APIs Agentic RAG Structured output Tool calling Speech-to-text PyTorch |
| Front end | React WebGL Chrome extensions |
| Shipping | Docker Nginx Linux Git Playwright Pytest CI |
- PlanForge: a constraint-satisfaction framework that generates apartment floor plans, with a visual solver that replays the backtracking tree.
- Relationship Analysis: a graph-based relationship analysis API. A team project from Star Academy, summer 2024.
- Audio-video recorder: Arduino and ESP32-CAM recording kept in sync to about ±10 ms, with offline transcription.
- Smart Monitoring System: ESP32-CAM monitoring with AI-assisted analysis.
- ASL to LCD: sign-language letters from a webcam onto a 16×2 LCD through an Arduino. The vision pipeline is adapted from Tanmay Jivnani's open-source project.
- DS Social Network: social-network APIs and the graph algorithms behind them.
- Coursework: AI Fundamentals, Machine Learning, Computational Intelligence.
Email is best: msmahdinejad@gmail.com. I'm also on LinkedIn and Telegram (Vertex Studio). There's more on the website.
