Evgenii Zakharov

Senior Full-Stack Engineer

Evgenii ZakharovSenior Full-Stack Engineerto@risen.dev • Remote worldwide (contractor or EOR), Yerevan (UTC+4) • https://linkedin.com/in/risenx •https://risen.dev • https://github.com/risenxxxPROFESSIONAL SUMMARYSenior full-stack engineer with 7 years in TypeScript, React and Node.js. I take products from idea toproduction and keep them running: the interface, the services behind it and the Kubernetes infrastructure,with the UI held to the same bar as the backend. At X5 Group I made food.ru (7.5M monthly users) fasteron mobile and raised its page-cache hit rate from 40–50% to 80–85%; at SberEducation I led a team ofthree frontend engineers. Creator of nestjs-zod (5.5M npm downloads a month). Now building Tolkly, an AIsupport chat that answers from a site's own docs.WORK EXPERIENCEX5 Group • Remote • 04/2024 - 09/2026food.ru recipe platform (7.5M MAU) at Russia's largest food retailerSenior Full Stack Engineer• Raised mobile Lighthouse Performance on key Next.js SSR pages by 15–25 points and cut shippedJavaScript from 1.1 to 0.7 MB through code splitting and tree-shaking.• Raised Redis page-cache hit rate from 40–50% to 80–85% after finding in Grafana that utm_* parameterssplit cache keys; cached pages return in 100–200 ms instead of a 700–800 ms SSR render.• Moved Keycloak auth to a server-side flow on Next.js API routes: pages render signed-in from SSR with nologin flicker, and tokens live in HttpOnly cookies instead of localStorage.• Built a Next.js BFF over several FastAPI services, partitioning cache keys by A/B bucket (Varioqub) soexperiments keep per-user variants without losing caching.• Cut data-mapping Sentry issues from 120 to 40 a month by tightening API contracts with runtime validation.SberEducation • Remote • 10/2021 - 03/2024EdTech arm of Sber, Russia's largest bank. Until 09/2022 via contractor Haiku Dev, then in-houseSenior Full Stack Engineer, Frontend Team Lead• Led a team of 3 frontend engineers for a year: mentored mid-levels through code review and pairing andset team standards for API contracts, accessibility and testing.• Designed versioned BFF contracts over upstream services with runtime validation and CI-generatedGraphQL types (Hasura), so breaking changes fail in CI, not in clients.• Built a mentor–mentee product from scratch: service architecture on NestJS, TypeORM and RabbitMQ,real-time communication, a Telegram bot and n8n automations.• Reached Lighthouse Accessibility 95+ on the payment widget and shared UI primitives, with consistentkeyboard and focus behavior across monetization paths.Winterfell Hotels • Remote • 10/2019 - 09/2021Internal systems for a hotel chainFull Stack Engineer• Built a hotel-management CRM used daily by front-desk staff: a drag-and-drop room × date booking gridwith server-side conflict checks, so concurrent edits never corrupt a reservation.• Integrated fiscal receipt printing end to end with retries and reconciliation, so no payment is left without areceipt, or a receipt without a payment.
PROJECTSnestjs-zod05/2022 - 10/2024 • npmjs.com/package/nestjs-zodOriginal author of the standard way to use Zod with NestJS: 5.5M downloads a month, 1,000+ GitHub stars;one schema drives validation, TypeScript types and OpenAPI docs. Handed to a new maintainer in 2024.Tolkly10/2026 - Present • tolkly.ioAI support chat that answers from a company's own docs and rules• Every statement of an answer cites its source page and is checked by code against a verbatim quote;questions the docs don't cover are refused and handed to a person.• On a 20-question eval over GrowthBook's docs: 15/15 covered questions answered correctly and all 5uncovered ones refused.• Replaced keyword handoff rules with a classifier: the decision to hand a chat to a person is right in 92 of 94cases across four businesses and five languages (keyword rules: 33 of 51).• Routed each message to Claude Haiku or DeepSeek by the site's line of business, the visitor's countryand device time zone, so visitors a provider excludes still get answers; every uncertain signal falls to thepermitted model.Gift Fight11/2023 - 10/2025Real-time PvP Telegram game built with a business partner; launched to ~1,500 players (~500 RPS at peak),later closed on the funnel data.• Built game analytics on Kafka (Redpanda) with batched writes to ClickHouse; 1M game events went throughwithout loss or duplicates.Tuned08/2026 - Present • tunedhq.ioA single home for your career history and resumes, founded and built alone in TypeScript: one completebase per direction, with a version for each job posting built from it. From first commit to live payments in afew weeks; private codebase.• Replaced a three-pass agent with one schema-validated LLM call after an eval on 60 real resume–jobpairs: cost per run $0.098 → $0.047, median time 56 → 38 s, runs below the untailored control 21% → 0%.• Gated every prompt and model change on an eval harness over 22 real resumes and 1,100+ postings; 50+runs across four providers moved tailoring to Haiku 4.5 at a third of Sonnet's price.• Designed the pipeline so personal data is redacted before any model call, every claim traces back to thebase resume, and page fitting and scoring stay in deterministic code.• Shipped an MCP server that lets users' own AI agents edit their resumes with scoped tokens.Frame Player07/2026 - Present • frameplayer.appVideo player for Windows and macOS in Rust, Tauri 2 and libmpv: casts 4K HDR to TVs without re-encodingand streams torrents during download; libmpv patched on macOS to render behind a transparentwebview.SKILLSProgramming Languages: TypeScript, C# (.NET), SQL, RustBackend: Node.js, ASP.NET Core (.NET 9), NestJS, Hono, GraphQL, Socket.IO, Keycloak, Microservices,Distributed systems, Bun, Elysia, oRPC, ZodFrontend: React, Next.js, Core Web Vitals, TanStack Router / Query, Tailwind, Radix UI, Vite, Astro
Data: PostgreSQL, EF Core, Drizzle, Redis, NATS, Temporal, BullMQ, RabbitMQ, Apache Kafka, Redpanda,ClickHousePlatform: Kubernetes, Docker, Pulumi, Terraform, Ansible, FluxCD, Helm, Cloudflare, Hetzner, DigitalOcean,GCP, Prometheus, Grafana, Talos, Cloudflare WorkersPractices: Testcontainers, CI/CD, AI-assisted development (Claude Code), GitLab CILanguages: English (full professional proficiency), Russian (native)AI / LLM: LLM pipelines, Structured outputs, Prompt engineering, Prompt caching, LLM evals, Anthropic API(Claude), OpenAI API, Model Context Protocol (MCP)