Zahar

Z A H A R G U S YAT I N
A I S O F T WA R E E N G I N E E R , A G E N T O R C H E S T R AT I O N
Remote, stable connectivity | [open contact info](look above in the "contact info" section)

EXECUTIVE SUMMARY

Engineer with 5.5 years of building production software, including systems I designed, shipped and then operated
myself rather than handing to someone else. Orchestrating agents is already how I work rather than a new way of
working I would need to adopt: I decompose a task into pieces an agent can actually do well, write the
specification and constraints it works inside, verify what comes back against tests and my own reading, and own
every line that ships. Nothing merges that I cannot explain. I introduced that process across my team with explicit
guardrails, and I went a step further and built the tooling underneath it, publishing an MCP server on NPM that
exposes structured, schema validated tools and domain knowledge so an agent applies the right pattern instead of
guessing. Applied LLM work in production covers structured output and tool use, prompt and instruction design
that makes a model reliable at a specific task, confidence thresholds and guardrails, and explicit fallbacks when
output cannot be trusted. I keep consequential logic in deterministic code that can be tested and treat model output
as something to check rather than believe. Backend in Python and Node.js over PostgreSQL with data modelling,
SQL and query optimization, queues and workers, Docker, CI and secrets handling. Comfortable in unfamiliar
territory: I have shipped in mobile, media pipelines, fintech and marketplaces, and I can design a thing, direct
agents through it and recognise when what comes back is wrong. English B2.

SKILLS

Agent orchestration and task Guardrails and confidence Reliable services, retries,
decomposition handling idempotency
Specification writing and Determinism boundaries in AI Queues, workers, data pipelines
constraints systems Docker, CI, secrets, access control
Verifying and owning agent output Python 3, FastAPI Logging, monitoring, alerting
Claude Code and coding agents SQL and data modelling Integration testing against real
(daily) PostgreSQL, indexing, query dependencies
MCP server development, NPM optimization Internal tools and developer
publishing Node.js, NestJS, TypeScript tooling
Applied LLM engineering System design, boundaries, failure Working across unfamiliar
Structured output and tool use modes domains

EXPERIENCE

ChallengeSoft Nov 2020 - Present
Lead Engineer, grew from Junior (2020) → Senior (2022) → Lead (2024)
Work through agents daily: decompose the task into pieces an agent handles well, write the specification and the
constraints it works inside, then verify the result against tests and review it as strictly as a colleague's pull
request. I own what ships regardless of who or what wrote it.
Introduced that process across the team with explicit guardrails, so delivery accelerated without the quality bar
moving: lint, type and test gates run on every change no matter the author, and nothing merges that the author
cannot explain.
Built the layer agents work through: an MCP server with schema validated tool contracts, plus repository
conventions and tooling so an agent starts with scoped context rather than guessing.
Shipped applied LLM features in production: structured schema validated output, tool calling, prompt and
instruction design tuned for a specific task, confidence thresholds and explicit fallbacks when the response

cannot be used.
Kept consequential logic deterministic and testable, treating model output as input to be validated rather than a
decision to be trusted, which is what made those features safe to ship.
Designed systems before writing them: where the boundaries go, what each service owns, and how it fails, then
built services in Python and Node.js over PostgreSQL with data modelling, migrations, indexing and query
optimization.
Built reliable, unattended processing: queues and workers, scheduled jobs, retries with backoff, idempotent
handlers and dead letter paths, with structured logging and alerting so failures are loud rather than silent.
Operated what I shipped: Docker containers, CI pipelines, environment and secret handling, role based access
control, and monitoring with triage after release.
Worked across unfamiliar territory repeatedly, from mobile and media pipelines to payments, identity
verification and marketplaces, learning each well enough to design it and to tell when a solution was wrong.
Led the team on architecture, standards, code review and mentoring.

Independent Projects 2021 - Present
Self-employed, own products and agent tooling
Authored and published an NPM package: an MCP server built on the Model Context Protocol SDK that gives
coding agents structured tools and domain knowledge with schema validated contracts.
Built and shipped an application live on the App Store with a 4.8 rating, covering account and transaction flows,
funding by card and bank, identity verification and encrypted key storage.
Build internal tools and prototypes regularly, which is where most of my agent driven workflow was worked out
before it went into team practice.

LANGUAGES

Ukrainian (native), Russian (native), English (Upper-Intermediate, B2)