AI-First Senior Software Engineer
Ukraine, Kyiv · On-Site · Full-Time
Company: Orbox Group - new AI-native product for the oil & petroleum business in Ukraine
Location: Ukraine, Kyiv - on-site, office in the city centre
Employment: Full-time, permanent
Seniority: Senior
Reports to: Tech Lead
Works with: Business Analyst, UI/UX Designer, QA Engineer, business owners & operations team
Mobilization reservation: Full reservation (deferment) from mobilization for male employees
Languages: English - Advanced (B2-C1); Ukrainian - fluent
The role
This is the role that turns a real offline business into production software - by leading AI agents, not by typing every line yourself.
This is a hands-on senior engineering role. Most of our code is written by AI coding agents against machine-readable specs. That makes senior engineering judgment more valuable, not less: agents write fast, but still pick wrong abstractions, miss edge cases, get money and inventory maths wrong and ship code that passes its own tests and breaks in production. You are the engineer who designs the system, sets the guardrails and owns what gets merged.
We operate in a fast-moving startup environment where speed and precision both matter. You must be opinionated about which engineering tasks AI does well today (boilerplate, refactors, tests, migrations, docs), which it does not (architecture, security boundaries, trade-off calls, debugging subtle production issues) - and how to combine both deliberately.
What you will own & do
AI-First Engineering
• Daily, deep, native use of AI coding agents for implementation from specs, refactoring, test writing, code review, debugging and documentation.
• Treat agents like a team you lead: clear specs in, reviewed code out - you own the context, rules files, guardrails and the review bar.
Set up the agentic delivery pipeline for the new product from day one: repo conventions, agent instructions, MCP tooling and CI checks that keep agent output safe to merge.
• Contribute to internal evals from the engineering side: define what good” looks like for AI-generated code and for the product's user-facing AI features.
Product Engineering
• Own the architecture from zero: service boundaries, domain and data model, APIs, environments - built for a real operational business with multiple sites and roles.
• Own features end to end: from spec & acceptance criteria to design, implementation, tests, deployment and monitoring on prod.
• Build operational surfaces: dashboards, inventory and stock movements, orders, pricing, documents, approvals, reporting and back-office tooling - including mobile-friendly flows for people in the field.
Integrations & Data
• Integrate with the existing world: accounting and ERP systems, Excel and legacy data, banks and payment providers, fiscalization and regulatory reporting.
• Connect the physical side of the business where relevant.
• Plan and run data migration from spreadsheets and legacy tools - reliably, idempotently, with proper error handling and audit trails.
AI & LLM Systems
• Build production AI features: LLM agents, tool calling, RAG, document processing (waybills, invoices, contracts), reconciliation, forecasting and anomaly detection - with grounding, fallbacks and cost control.
• Engineer the non-deterministic layer: prompt versioning, structured outputs, guardrails against hallucinations and prompt injection, role-safe access to data.
• Instrument AI behaviour with tracing, logging and metrics - treat AI output as something to be measured, not trusted.
Engineering Quality & Ownership
Own the merge bar: review agent and human PRs, enforce architecture, security and performance standards.
Write automated tests and CI that make agent output safe to ship; partner with QA on test strategy and release readiness.
• Feed spec & story defects back upstream: when the work item is wrong or ambiguous, it gets fixed at the spec layer with the Busi
