ReadMe.md

closetema

Matthew Demidoff

Software engineer. I build and run production systems end to end: Node/TypeScript and Python in the application layer, Rust where it earns its place, and the infrastructure underneath them.

Currently shipping for three businesses across Russia, Serbia and the EU. Open to software engineering internships.

Main contact: @bnecko - [email protected] - bottleneck.cc

CV (PDF)


Experience

Software engineer - AVANTDEMI (Serbia, property agency) - 2026 to present

Internal tooling and the public site for a Belgrade-based property agency.

Software engineer - Bottleneck powerSource Group - 2025 Q3 to present

Identity for everything the group runs: auth.bneck.com is an OAuth 2.0 / OIDC provider with WebAuthn, Telegram 2FA and session management. Open source: github.com/bnecko/auth.

Software engineer - commercial real-estate group (Russia, warehousing) - 2025 Q2 to present

Built the tooling around the brokers' CRM and reworked the sales workflow end to end: listing-to-demand matching, an AI co-pilot for the brokers, and email/Telegram integrations into one pipeline.


Projects

Bottleneck OAuth Platform - the identity provider behind my projects. OAuth 2.0 / OIDC, WebAuthn, Telegram 2FA. Source on GitHub: bnecko/auth. Open to collaboration.

MiniComm on Telegram**

Everything else is on my GitHub


Infrastructure and tooling

Google Cloud Platform - production workloads.
Yandex Cloud - for Russian clients, where latency and local data-residency rules make it the right choice.
Cloudflare - static hosting, DNS and domain management, DDoS protection.
Docker - multi-application hosts, the deployment target for most of the above.
Rust - smaller tools where the performance or the single-binary distribution is worth it.
Web3 - smart contracts on Solidity for TRC-20 and ERC-* networks.

AI is part of most of what I ship, and the provider is a constraint, not a preference: Russian clients need a provider that complies with Russian law, privacy-bound clients need local models, and cost-sensitive workloads need a price/token ratio that Western frontier models do not offer. I have also trained models myself*, both fine-tuning and from scratch, on owned and rented compute.

Providers I have shipped against
  • OpenAI, Anthropic - general enterprise work
  • Yandex AI - Russian legal compliance
  • Local LLMs - privacy-bound clients
  • Gemini - audio workloads
  • DeepSeek, Qwen, Z.ai GLM, Kimi - cost-sensitive workloads

Languages

Serbian, Russian, English - B2 to C1, fluent in professional use.


Links & Contacts

Main Account: @bnecko
Discord: bnecko
Signal: larp.33

Mail: [email protected]
Websites: bottleneck.cc and bneck.com


* - Both fine-tuning and training from scratch on owned and rented compute (mainly GCP)

** - MiniComm is currently in testing, and registration requires admin approval. The standard process will be rolled out later. You can test the open beta at app.bneck.com

Noticed something wrong AND/OR got any suggestions? - Contact me in any form and I will reach out for you.

bayan