Software Engineer - AI

Custral•New York City, NY

6d•Onsite

About The Position

Custral is building a new kind of workspace that goes beyond traditional CRMs. Unlike other tools that stop at the closed deal, Custral covers the entire customer lifecycle, including implementation, support, renewal, and product feedback, all within a single, integrated platform. This eliminates the need for multiple disparate tools and prevents teams from having to adapt their processes to software limitations or resort to spreadsheets for critical data. Custral's AI agents are deeply integrated into the data model, offering advanced scoping, tool gating, and model selection across OpenAI and Anthropic. Stella, the AI assistant, operates on the same customer records as the rest of the product and always seeks confirmation before sending information. This role focuses on developing the core AI agent functionality within Custral, ensuring these agents are not merely add-ons but integral components with sophisticated capabilities. The position involves owning the runtime infrastructure, including the tool layer, context assembly from Postgres and OpenSearch, error tracing, and performance evaluation. A key aspect of this role is building robust evaluation frameworks and regression suites to objectively measure and improve agent quality, moving beyond subjective judgment to evidence-based decision-making. This is a full-stack position, encompassing both front-end development with React for agent surfaces and back-end development with Node and Express for the runtime. Initial projects include developing an agent evaluation harness for evidence-based shipping of prompt and model changes, and enhancing Stella's capabilities with new tools, improved context assembly, and expanded interaction surfaces.

Requirements

Deep fluency in TypeScript and comfort across the full stack.
Experience shipping LLM features to real users and maintaining them as models evolve.
Experience building evaluations or regression suites for non-deterministic systems.
Understanding of retrieval and context assembly as engineering problems.
Experience with Postgres and a search-oriented database like OpenSearch or Elasticsearch.
Ability to research independently and drive work forward without constant direction.
Strong candidates ship real functionality without cutting corners, explain their tradeoffs, and leave the system better than they found it.

Nice To Haves

Agent frameworks and MCP
Streaming UIs
Redis for queues
Cost and latency tuning at volume

Responsibilities

Own the runtime infrastructure for AI agents, including the tool layer, context assembly from Postgres and OpenSearch, and error tracing.
Build evaluation frameworks and regression suites to objectively measure and improve AI agent quality.
Develop agent surfaces using React.
Develop the runtime using Node and Express.
Enhance Stella's capabilities with new tools, improved context assembly, and expanded interaction surfaces.