Personal & open source

Kelid (کلید)

Personal project · in development 2026

A custom iPhone keyboard with a clipboard manager, live resizing and Persian word prediction that learns from you, running entirely on the device with no network access.

  • Clipboard history with pinning, search, snippets and text expansion, sensitive-content filtering and a Face ID lock.
  • Live drag-to-resize, one-handed mode and separate size profiles for portrait and landscape.
  • Its own memory-mapped language model (a trie plus n-gram tables, Stupid Backoff scoring and a typo model that knows Persian homophones), blended with a personal model that learns on the device.
  • A Python and DuckDB pipeline that builds the Persian and English n-gram data from Wikipedia and word-frequency lists.
  • Swift
  • SwiftUI
  • iOS
  • SQLite
  • Python
  • DuckDB

macsmartcleaner

Personal project · open source (MIT) 2026

A free, open-source terminal app that shows what is filling up macOS "System Data" and cleans it safely, with an animated interface and no dependencies.

  • Smart Clean clears caches and junk but skips the apps you have open, and Deep Scan checks 92 known junk sources plus anything unusually large.
  • Space Lens, an uninstaller that removes app leftovers, startup-item management, one-key optimisations and a live CPU, GPU, memory and disk dashboard.
  • Safe by design: personal data and system folders are hard-blocked, everything asks first and removals go to the Trash. No telemetry.
  • Auto-Protect installs an hourly LaunchDaemon that keeps the Spotlight index and free disk space in check.
  • Python
  • curses
  • macOS
  • CLI

Quaderno

Personal project · open source (MIT) 2026

A self-hostable study notebook for language learners. Class handouts in any format become annotatable PDFs, you can write your own pages between them, and it shows how many hours you really studied.

  • Converts PDFs, Word files, slide decks and images to PDF on upload, with OCR for scanned handouts, and always keeps the original file byte-identical.
  • An annotation layer with highlights, a freehand pen with pressure and palm rejection, threaded comments and your own note pages inserted anywhere, exported as flattened or layered PDFs.
  • Study tracking with a live timer, weekly goals, charts, a year heatmap and streaks, plus SM-2 spaced-repetition vocabulary review with Anki export.
  • An installable, offline-tolerant PWA built to run on a €5/month VPS, with a separate worker for conversion and OCR and PostgreSQL doubling as the job queue.
  • Next.js
  • TypeScript
  • PostgreSQL
  • Prisma
  • PWA
  • Docker

Work

GetSpoused dating app

Full Stack Engineer · Contract 2024 – 2025

A production dating app delivered end to end as the sole engineer, with React Native clients for iOS and Android, a React marketing site and an Express.js back end.

  • Built the back end on MongoDB with Redis caching and RabbitMQ for matching, notifications and media processing.
  • Designed the matching and messaging data model.
  • Shipped the whole product on roughly 20 hours a week.
  • React Native
  • React
  • Express.js
  • MongoDB
  • Redis
  • RabbitMQ

AI customer-service platform

System Architect & Back-end Developer · Wormpost 2024 – 2025

System architecture and back end for Wormpost's AI customer-service product, from the load-balanced microservices to the agent pipeline that handles customer conversations.

  • Split the platform into microservices behind a Kong API gateway, with load balancers in front of them.
  • Used Kafka for event streaming and queueing, Redis for caching and rate limiting, and PostgreSQL and MongoDB as the databases.
  • Built the AI agent pipeline, including retrieval and tool-calling paths, and extended the architecture to a second product line.
  • Go
  • Django
  • Express.js
  • Kafka
  • Redis
  • Kong
  • AI agents

Solana wallet analytics

Full Stack Engineer · Contract 2024

A crawler that analysed more than 10,000 Solana wallets for win rate and realised profit, then ranked them as copy-trading candidates.

  • Worked around strict third-party API rate limits by sharding the workload across 40+ parallel workers on Kubernetes, with back-off and checkpointing.
  • Persisted and de-duplicated results so repeated runs were incremental rather than full re-crawls.
  • Kubernetes
  • Crawling
  • Rate limiting
  • Solana

English Bot

Software Engineer · Protein Team 2023 – 2024

A Telegram-based English tutoring platform that first matches learners of the same level and falls back to a self-hosted LLM conversation partner.

  • Self-hosted a 14B model on Ollama instead of paying per token for OpenAI, cutting inference cost to infrastructure cost.
  • Wrote the matching engine in C++ for throughput and connected it to the rest of the system through API gateways in a microservice layout.
  • Used Redis and Kafka for load balancing and queueing across the conversation and matching services.
  • LLM
  • Ollama
  • C++
  • Redis
  • Kafka
  • Telegram

ECG analysis platform

System Designer & Full-stack Developer · Zistel 2022 – 2023

Full-stack work on an ECG analysis platform, with Express.js services on PostgreSQL and Angular and Vue front ends.

  • Built clinician-facing review tooling and the APIs behind it, including role-based access to patient records.
  • Express.js
  • PostgreSQL
  • Angular
  • Vue
  • Healthcare

Telecardiology platform

System Designer & Full-stack Developer · Zistel 2021 – 2022

System architecture and back end for a subscription telecardiology service that takes readings from a heart-analysis device, applies the research team's formulas and routes the results to a cardiologist.

  • Cut the manual review load per patient by automating the path from device reading to cardiologist.
  • Used RabbitMQ for asynchronous processing and MongoDB for storage.
  • Built the scheduling flow for follow-up calls, meetings and chat.
  • Go
  • Express.js
  • RabbitMQ
  • MongoDB
  • Healthcare

Rasad Bazar

System Architect & Back-end Engineer · Meghyas 2019 – 2021

A stock-market analysis platform for the Tehran exchange that turned Telegram signal channels into buy and sell recommendations.

  • Designed the architecture to handle 10,000 users at the same time.
  • Ran a fleet of Telegram bots that recorded every message and forwarded it to a sentiment-analysis service, which scored each stock.
  • Used Kong as the API gateway, with RabbitMQ, Kafka and Redis for message flow and caching.
  • Telegram bots
  • Sentiment analysis
  • Kafka
  • RabbitMQ
  • Kong

University

MSc thesis: trust management with negative requirements

MSc Computer Science · Università degli Studi di Milano 2026

My master's thesis in Computer Science at Università degli Studi di Milano.

  • Research
  • Trust management