Experience

the roles and the miles behind them

./turtle --drive

Lead Software Engineer (Python Backend)

Bemo — TON liquid staking
Jul 2024 — Present · Remote · part-time since Jan 2026
  • Tech lead in a cross-functional team of up to 8 with 3 direct reports; ran interviews and hired 2 engineers; mentored 3 (one grew to senior, one from junior to middle).
  • Merged two staking oracles into one deposit service, since neither alone reached the validator stake minimum; shipped via replay tests and a shadow run with 0.00% divergence from legacy, putting ~$850k of idle stake back to work.
  • Took over an unowned legacy backend: fixed lost on-chain events overnight before the next validation cycle, then moved event ingestion to interchangeable sources (own liteservers, toncenter, dton) with lag alerts.
  • Resolved a rewards-accrual incident: replaced the event source that dropped transactions, published a formula users could verify, and built a calculator that replays every balance from on-chain history.
  • Operated and upgraded 10+ FunC contracts in production; built the admin multisig console (FastAPI + React, wallet-signature login) and Python services that build, sign, and verify contract messages.
  • Introduced Sentry after a one-month self-hosted trial and added on-chain invariant alerts; held 99.9% validator uptime on the public TON dashboard.

Gravity, liquid staking on the TAC chain

  • Built a TON DeFi liquidity indexer over 17 contracts on 6 protocols (STON.fi, DeDust, EVAA): asyncio worker per protocol, saved cursors, duplicate-safe ingestion, lag monitoring.
  • Turned the XP rewards service into a multi-program engine and shipped a public APY API computed from on-chain mint rates (web3.py, rate limiting, historical backfill); wrote the DefiLlama TVL adapter.

Python (FastAPI, Pydantic, SQLAlchemy, APScheduler, pytoniq, web3.py, pandas), PostgreSQL, Redis, React, Docker, Sentry, GitLab CI

Senior Python Backend & AI Engineer

Independent contractor — clients under NDA
Jan 2026 — Present · Remote

Client: B2B electronics wholesaler, 1,400 suppliers

  • Built an LLM service that turns free-text supplier price lists into 18-field SKU records for price comparison: 4.9M price lines from 1,400 suppliers in its first 5 weeks, ~3 SKUs per second on a self-hosted Qwen2.5-7B (vLLM, regex pre-pass).
  • Raised the share of rows free of systematic errors, audited against the client's production data, from 87.6% to 99.5% over 7 releases gated by golden, held-out, and prompt-injection eval sets in CI.
  • Architected an internal AI-agent platform, in pilot and rolling out to all 50 employees across 10 departments: each employee session runs a Claude Code agent in a hardened Kubernetes sandbox pod, with per-user, per-team, and per-role quotas and read-only CRM data with personal fields removed.

Client: B2C sales company, 300+ employees, sales run through Telegram

  • Took over a single-developer finance bot and grew it into a 9-service platform on Kubernetes with GitOps, Sentry, and alerting, used by sales, finance, and management teams.
  • Migrated the system of record from MongoDB to PostgreSQL module by module behind a storage facade; a diff harness on live data confirmed parity on 316k payment records.
  • Replaced a Google Sheets P&L with an in-house reporting app; fixed a currency defect in the legacy revenue report that silently dropped foreign-currency payments, verified by an exact SQL reconciliation.
  • Consolidated ~30 copy-pasted Telegram marketing bots into one multi-tenant service where a new bot is a database row; migrated ~955k users with zero downtime in ~2 GiB of memory.
  • Built a subscription AI-psychologist bot (streaming OpenAI replies, crisis detection before every LLM call, recurring payments) and hardened a contractor-built Telegram CRM used across 134 sales accounts.

Python (FastAPI, aiogram, SQLAlchemy, Alembic, Taskiq, FastStream), vLLM, Claude Agent SDK, OpenAI, PostgreSQL, MongoDB, RabbitMQ, Redis, Kubernetes, FluxCD, Vault, OpenTelemetry, GitHub Actions

Python Backend Developer

My Path — AI products for HR-tech and computer vision
May 2023 — Apr 2024 · Remote
  • Built an internal multi-tenant LLM gateway: per-client credits, batch calls, structured JSON output, token accounting (FastAPI, LangChain).
  • Co-built a resume-parsing service: PDF to structured profile via layout-based section splitting, DeepPavlov BERT NER for names, and GPT-3.5 extraction, with every value checked against the source text to flag hallucinations.
  • Designed a filter language for recruiters (nested AND/OR trees of 26 operations compiled to SQL) to shortlist parsed candidates by vacancy requirements: experience, skills, languages.
  • Built backend and ETL stages for a computer-vision pipeline monitoring construction sites: camera-archive extraction, Redis task queue, per-zone work-hour reports (YOLOv8 and face re-ID on Triton).

Python (FastAPI, Flask, SQLAlchemy, Celery, LangChain), PostgreSQL, Redis, Docker, Kubernetes, Helm, GitLab CI

Python Backend Developer

Kortechs — Altaport heliport-reservation marketplace
Apr 2022 — Apr 2023 · On-site
  • Built the Altaport backend from scratch in a team of 3 backend engineers: Django/DRF marketplace on GKE with a 10-state booking workflow (django-fsm), role-based transitions, and an audit log.
  • Integrated Stripe Connect (operator onboarding, payments, refunds, payouts, monthly invoices) plus WhatsApp (Twilio) and email (SendGrid) notifications.
  • Put TimeKit behind a pluggable reservation interface and replaced it with an in-house scheduler; contributed reschedule and cancel support to the upstream python-timekit library.
  • Cut CI/CD time from 20 to 5 minutes with parallel test runs (pytest-xdist), Kaniko layer caching, and multi-stage images; built a Slack bot that cuts releases and triggers GitLab pipelines from chat.

Python (Django, DRF, django-fsm, Celery), PostgreSQL, Redis, Stripe, Docker, Kubernetes (GKE), GitLab CI

Software Engineer

NiceCode — web studio
Jan 2020 — Apr 2022 · On-site
  • Built a meal-plan backend from scratch (Django, Celery): splits daily calorie and macro targets across meals and picks allergen-free recipes from 3,800+ scraped recipes.
  • Built scraping services, a Telegram server-monitoring bot (aiogram, paramiko), Prometheus and Grafana business metrics, and GitHub Actions CI/CD.

Python (Django, DRF, Celery, aiogram, Selenium), PostgreSQL, Redis, Prometheus, Grafana, Docker, GitHub Actions

← back to ~/ruslan-korneev