rvzd

Lev Zavodun

Backend engineer. Go and TypeScript, LLM ops and agentic pipelines, distributed systems.

Tbilisi, Georgia. Remote, EMEA

Summary

Backend engineer (4.5+ years, Go primary) who builds and operates LLM systems as production infrastructure: multi-model pipelines with validation layers, empirical model evaluation, observability and cost/latency/quality trade-offs. I own systems end-to-end — architecture, implementation, deployment, monitoring. Smart contracts I authored passed a CertiK audit with zero critical/medium findings. Ranked #13 on the TON Builder leaderboard; in the ecosystem since January 2022. Looking for platform / applied-AI roles where other engineers depend on what I build.

Experience

Blockchain Developer, MetaLamp

Aug 2024 – Jul 2026

full-time, remote

  • Designed and shipped TON smart contracts (FunC / Tolk / Tact) end-to-end: architecture → mainnet → monitoring. Latest system — distributed staking with per-user contracts — passed CertiK audit with 0 critical / 0 medium findings.
  • Built the surrounding infrastructure for every contract system: indexers, backends, monitoring, typed client wrappers (Go, TypeScript, PostgreSQL, Redis).
  • Identified, disclosed and remediated an unauthorized-transfer vulnerability (unverified DEX callback) in a client’s production contract.
  • Worked in a fully remote, async team with clients across time zones; owned features from spec through post-release support.

Independent Blockchain Developer, TON ecosystem

Jan 2022 – Aug 2024

freelance, remote

  • Contributed merged pull requests to the official TON documentation; bounty contributions listed in the TON Society Bounties Hall of Fame (as Revuza).
  • Ranked #13 on the TON Builder leaderboard (Platinum tier) and #39 Contributor; in the ecosystem since its first open year.
  • Built and optimized FunC contracts for ecosystem projects; occasional backend consulting (Go, contract tooling) continues to date.

Projects

Multi-model LLM pipeline + evaluation harness
Orchestrator / worker / cross-family checker architecture to mitigate model self-preference bias; all mutable actions gated by human approval, capability bounds enforced at the harness layer. Validated with an 80-session benchmark (Jaccard deduplication, cross-run overlap analysis) that drove model-selection and cost decisions.
rvzd
Self-hosted infrastructure monitoring with an AI incident layer. Go hub and agents: agent-initiated WebSocket command tunnel (no inbound ports), TLS via ACME, revocable per-agent tokens; three-tier metric compaction; checksum-verified self-update with automatic rollback, binaries on Cloudflare R2. Tool-using LLM triage over read-only diagnostics; fixes run only after human approval, followed by scheduled re-checks and a post-run judge. HMAC IP pseudonyms and secret redaction on everything sent to the model. Prometheus and Alertmanager ingestion. Backend-less React UI with Telegram OTP session 2FA.
anypocket
Browser extension that saves pages to a local Anytype as bookmarks with notes. Firefox and Chromium builds, five languages, no server and no analytics.
Вахтёр
Telegram Business gatekeeper bot: LLM triage agent with human-in-the-loop escalation. TypeScript, Bun, grammY.
Video fact-checking pipeline
Multi-agent claim extraction and verification with cross-family model checks and configurable domain profiles.
tt-easy
Multi-node VPN management panel: Go backend, gRPC+mTLS agent control plane, React frontend, Docker-based routing chains.

Skills

Languages
Go (primary), TypeScript, Python (working), FunC / Tolk / Tact, SQL
LLM / AI
Multi-model orchestration, evaluation & benchmarking, agent harness design, MCP, cost / latency / quality trade-offs, human-in-the-loop controls
Systems
gRPC, mTLS, distributed systems, observability & metrics pipelines, PostgreSQL, Redis, Docker, Linux, Cloudflare R2
Security
Smart-contract security (CertiK-audited), responsible disclosure — credited in Anytype’s security contributors; auth design (revocable per-agent tokens, OTP session 2FA)

Languages

English
C2 (EF SET 78/100, 2026)
Russian
native
Japanese
beginner

Education

Bachelor’s degree — Ural Socio-Economic Institute (branch of the Academy of Labor and Social Relations, Moscow)