Agentic Software Engineer (Full Stack)

About the role

Specright is a purpose-built, patented platform for specification management that helps companies digitize and map their supply chains. It is looking for an Agentic Software Engineer (Full Stack) to ship reliable, user-facing AI agent products to production quickly. This is not a research role: you will own production systems end-to-end, from agent architecture and code to evaluations and operations. The position is fully remote.

Key responsibilities

• Design, develop and deliver agentic products end-to-end, including chat-based interfaces, AI workflow tools and internal copilots used by real users in production.
• Use agentic coding tools daily (e.g. Claude Code, Codex, Cursor) to speed delivery while keeping high code quality.
• Architect agent internals: tool calling, multi-step orchestration, memory and state management, retrieval pipelines, permissions models and failure recovery.
• Build and maintain agent evaluation infrastructure: regression suites, scenario-based tests, golden traces and offline and online metrics.
• Instrument agents with observability: distributed tracing, prompt and version tracking, tool-call telemetry and cost/latency monitoring.
• Own production readiness: rollout plans, QA, guardrails, incident response and continuous improvement of deployed agents.

Required profile

• Proven full-stack engineering ability, having shipped production software end-to-end.
• Hands-on experience building LLM-powered agents that work in production, owning at least 1-2 agentic features or products through deployment and operation.
• Familiarity with agent evals and tracing beyond model benchmarking, including trace-based debugging and regression detection.
• Solid engineering fundamentals: testing, debugging, security basics, performance optimization and reliability.

Nice-to-haves

• TypeScript/Node.js and/or Python, with experience in modern web stacks and startup pace.
• Agent tooling ecosystems such as OpenTelemetry, LangSmith, Datadog or Honeycomb.
• RAG architectures: vector databases, indexing and chunking, retrieval evaluation.
• Tool APIs for agents: structured output schemas, idempotency, rate limiting and error handling.