Hey, I'm Arash 👋 a backend engineer who likes building things that have to keep working: payments, APIs, infrastructure, and lately a lot of AI agents.
- 🎮 Most recently I led the backend of a live gaming platform with real money flowing through it: payments, a creator marketplace, a game launcher, and the tests and monitoring that kept it all running.
- 🤖 I've been building AI features in production: sandboxed agents with MCP tools, real-time voice, and memory for long conversations.
- 🌐 Before that I ran my own fleet of Linux servers and wrote networking tools in Go, which is where sluice came from.
- 🔬 On the side I tinker with machine learning: hypergraphs, reinforcement learning and a bit of computer vision.
💬 Curious about my work? Ask my AI at ask.devarash.icu. It answers from my own notes and shows where each answer came from.
|
Framework-free RAG assistant that answers questions about my work with cited sources. A relevance gate calibrated from measured similarity, injection-resistant prompts, Postgres rate limits on HMAC'd IPs, and an evaluation suite to keep it honest. |
TCP tunnelling in Go: multiplexed reverse tunnels over TLS with certificate pinning, so many logical connections share one secured link, with Prometheus metrics for throughput and connection health. |
|
Real-time crypto market microstructure: live order books from several exchanges, normalised into one model and kept healthy through disconnects, rate limits and format changes. |
Zero-dependency network and security toolkit: X.509 certificate tooling, AES-256-GCM with hand-written PBKDF2, RSA-OAEP, network diagnostics, and authorised load testing. Fully automated semantic-version releases. |
|
Self-hosted, privacy-first cloud storage on any S3-compatible backend, plus an Android client in React Native with a native Kotlin SSH/SFTP module. Built under ZeroNode Labs. |
Hypergraph community detection: Node2Vec, DeepWalk and hyper2vec embeddings, k-means clustering, then h-Louvain refinement, evaluated against EC-Louvain and h-Louvain on modularity and convergence. |
More projects
- Reinforcement Learning Playground: PPO, DQN and SAC agents across six Gymnasium environments with custom reward shaping.
- Facial attribute classification: PyTorch model combining ResNet-18, a bidirectional LSTM and multi-head attention.
- DirectPortForward-Go: high-performance TCP port forwarder, the predecessor to sluice.
- e-Vision: online-shop builder with CMS, CRM, a React admin panel and a Next.js page builder.

