Home › Career at SDK.Finance › AI Workflow / Agent Systems Engineer

AI Workflow / Agent Systems Engineer

14. 09. 2026

About SDK.finance

SDK.finance provides software infrastructure for companies building digital finance products. We are transforming our internal operations and product development workflows using AI engineering, agentic systems, retrieval, evaluation, and controlled automation.We are looking for a hands-on engineer who can build the technical systems that make this transformation real.

About the Role

The AI Workflow / Agent Systems Engineer will design and build production AI workflows, agent harnesses, RAG systems, tool integrations, evaluation pipelines, and internal automation systems for SDK.finance.This role is for a strong software engineer who understands that reliable AI systems require much more than calling a model API. You will build the surrounding infrastructure: context preparation, retrieval, structured outputs, tools, permissions, state, retries, observability, evaluation, approval flows, and recovery after failures.

What You Will Do

  • Build production-grade AI workflows for SDK.finance business and engineering processes.
  • Implement RAG systems over internal documents, product knowledge, CRM data, support materials, technical documentation, and code-related knowledge.
  • Build structured-output pipelines using JSON schemas, Pydantic, Zod, or similar validation approaches.
  • Design and implement tool-calling systems that allow AI agents to safely interact with APIs, databases, documents, internal systems, and developer tools.
  • Build agent execution loops, workflow orchestration, model routing, task state, retries, cancellation, checkpoints, and approval gates.
  • Implement evaluation harnesses for AI workflows, including deterministic checks, LLM-as-judge workflows, regression datasets, and negative test cases.
  • Track and optimize latency, token usage, cost, error rates, retrieval quality, and task success rates.
  • Design observability for AI systems, including traces, logs, model calls, tool calls, retrieved sources, decisions, and final outcomes.
  • Help build internal AI tools for sales, product, support, delivery, documentation, and engineering teams.
  • Work closely with product and business stakeholders to turn real workflows into reliable technical systems.

What We Are Looking For

  • Strong backend or full-stack engineering experience.
  • Strong Python and/or TypeScript skills.
  • Experience building production systems with APIs, databases, queues, async execution, and observability.
  • Hands-on experience with LLM APIs and AI application development.
  • Practical knowledge of RAG, embeddings, vector search, hybrid search, reranking, chunking, citations, and document processing.
  • Experience with structured outputs, function calling, schema validation, and tool integration.
  • Understanding of AI evaluation: test sets, regression cases, negative examples, judge models, and quality metrics.
  • Ability to debug AI systems across multiple layers: model behavior, prompt, retrieval, tool call, data source, workflow state, and business logic.
  • Good judgment around security, permissions, tenant isolation, sensitive data, and prompt injection risks.
  • Ability to write clear technical documentation and explain design trade-offs.

Nice to Have

  • Experience with LangGraph, LlamaIndex, LangChain, OpenAI APIs, Anthropic APIs, Azure OpenAI, pgvector, Postgres, Temporal, OpenTelemetry, vLLM, or MCP.
  • Experience with fintech, payments, ledgers, compliance, or regulated business workflows.
  • Experience building internal developer tools or coding-agent workflows.
  • Experience with evaluation platforms, AI observability tools, or model gateways.
  • Experience with document processing, OCR, PDF parsing, or knowledge-base automation.

Success in This Role Looks Like

  • SDK.finance has reliable internal AI workflows that are used by real teams.
  • AI outputs are structured, validated, traceable, and connected to evidence.
  • AI systems handle missing data, ambiguous inputs, permission limits, and partial failures safely.
  • The company can measure quality, cost, latency, and business usefulness of AI workflows.
  • AI systems become maintainable software, not fragile prompt experiments.