软件工程师 — AI 代理平台 / Polyant
Software Engineer — AI Agent Platform / Polyant
Exelab 和 Polyant
Exelab 围绕真实业务需求构建定制化的 AI 驱动解决方案。我们以小团队形式工作，并在设计和构建软件的过程中直接使用 AI。
你将加入开发 Polyant 的团队，Polyant 是我们开源的、可自托管的 AI 代理运行时。它通过 OpenAI 兼容 API 和管理界面，整合了记忆、业务文档检索、工具和插件执行、消息通道、权限、追踪和成本控制等功能。
申请前请先了解产品：
文档：
代码：
职位描述
你的主要职责是开发 Polyant：构建功能、提升可靠性并随时间推移帮助塑造产品。你将在整个平台上工作，从管理界面和运行时到数据库和发布流程，与团队协作，对定义的产品领域负责。
此职位适合有丰富经验交付和维护生产软件的工程师，能够将模糊的需求转化为明确的方案，解释架构权衡，并将变更推进到发布阶段。责任包括维护、文档编写以及从失败中学习。
研究和实验是工作的一部分。你将测试关于代理行为、检索、记忆和评估的想法，然后将有前景的结果转化为可维护的产品功能。重点在于软件和代理工程；模型训练、微调和 MLOps 不在此职位范围内。
你会做以下事情
- 从数据模型和 API 到用户界面、测试和文档，端到端地设计和交付功能。
- 在你负责的领域内做出技术决策，并明确权衡用户价值、实现难度、可靠性和维护成本。
- 开发代理功能，如工具执行、上下文管理、检索和记忆，并明确处理故障模式。
- 通过合理的授权、数据隔离、输入验证和仔细审查来保护多租户平台。
- 构建回归测试和代理评估；使用追踪、延迟和成本测量来评估变更。
- 通过 CI/CD、安全的数据库迁移、部署实践和对生产问题的调查来确保发布可靠。
- 有意识地使用 AI 编码代理：提供上下文和限制，将工作拆分为可审查的变更，并验证生成的代码是否符合预期行为。
- 提出产品改进方案，并分享实验、决策和发现结果
查看英文原文
About Exelab and Polyant
Exelab builds custom, AI-powered solutions around real business needs. We work in small teams and use AI directly in how we design and build software.
You will join the team building Polyant, our open-source, self-hosted runtime for AI agents. It brings together memory, retrieval over business documents, tool and plugin execution, messaging channels, permissions, tracing and cost control, behind an OpenAI-compatible API and an administration interface.
Explore the product before applying:
Documentation:
Code:
The role
Your primary focus will be Polyant: building features, improving reliability and helping shape the product over time. You will work across the platform, from the administration interface and runtime to the database and release pipeline, taking ownership of defined product areas in collaboration with the team.
This role suits an engineer with solid experience shipping and maintaining production software, who can turn an ambiguous need into a scoped proposal, explain architectural trade-offs and follow a change through to release. Ownership includes maintenance, documentation and learning from failures.
Research and experimentation are part of the work. You will test ideas around agent behaviour, retrieval, memory and evaluation, then turn promising results into maintainable product capabilities. The focus is software and agent engineering; model training, fine-tuning and MLOps are outside this role's scope.
What you will do
- Design and deliver features end to end, from the data model and APIs to the user interface, tests and documentation.
- Own the technical decisions within your areas and make explicit trade-offs between user value, implementation effort, reliability and maintenance cost.
- Develop agent capabilities such as tool execution, context management, retrieval and memory, with clear handling of failure modes.
- Protect a multi-tenant platform through sound authorization, data isolation, input validation and careful review.
- Build regression tests and agent evaluations; use traces, latency and cost measurements to assess changes.
- Keep releases dependable through CI/CD, safe database migrations, deployment practices and investigation of production issues.
- Use AI coding agents deliberately: provide context and constraints, split work into reviewable changes, and verify generated code against the intended behaviour.
- Propose product improvements and share experiments, decisions and findings with the team.
Requirements
Essential experience and skills
- A track record of delivering production software and maintaining it over time, with examples of decisions and outcomes you can explain.
- Strong TypeScript and Node.js skills, plus practical React experience and the ability to work across backend and frontend.
- Experience with relational databases, API design, authentication and authorization.
- Sound software design, debugging and refactoring skills, supported by automated tests and code review.
- Practical experience with LLM APIs and agent workflows, including tool calling, structured output, context limits and the impact of cost and latency.
- An evidence-based approach to reliability: you can explain how you test uncertain behaviour and investigate failures.
- Practical use of AI coding tools, with the ability to critically review their output.
- Experience with Git, CI/CD and deploying software; willingness to take responsibility for its behaviour after release.
- Clear written communication, fluent professional Italian and working English. Based in Italy.
Useful additional experience
- Next.js, NestJS, Drizzle ORM, PostgreSQL/pgvector, Docker or AWS CDK. These are parts of our stack; prior mastery of every tool is not required.
- Retrieval systems, embeddings, hybrid search, chunking or reranking.
- Agent evaluation, tracing, context compaction, lifecycle hooks, guardrails or MCP.
- Multi-tenant or self-hosted products, open-source contributions or developer-facing tools.
How we work and what you can expect
- A small team where your work contributes directly to a visible product and its technical direction.
- Technical collaboration, regular feedback and knowledge sharing as you build deeper expertise in agent systems.
- Room for focused experimentation, with a clear path from a hypothesis to a measured result and a product decision.
- A remote working environment in Italy, with ownership scoped to clear priorities and outcomes.
Salary range: 40-50 K
Originally posted on Himalayas