About the role
Responsibilities:
- Build real things. You'll write production code across the full stack: React frontends, Python/Node backends, ClickHouse queries, Kafka event pipelines, LangGraph agent workflows, and YAML-driven process definitions. Not prototypes. Not demos. Stuff that runs in production for real companies moving real cargo across real oceans.
- Talk to enterprise clients. You'll sit across from logistics managers, freight forwarders, and operations leads. You'll listen to their messy, exception-filled processes 6 approval levels, 4 document types, 3 systems that don't talk to each other and turn that chaos into structured workflows they can actually use. This isn't optional. It's the job.
- Think AI-first. Every problem you encounter, your first question should be: Can an agent handle this? Not: " Let me build a CRUD app. You'll design agents that extract fields from Bills of Lading, validate invoices against purchase orders, auto-route approvals based on business rules, and flag exceptions before humans even notice them. You'll work with LLMs, RAG pipelines, document parsers, and agent orchestration frameworks daily.
- Own problems end-to-end. You won't hand off a spec and wait. You'll identify the problem, design the solution, build it, configure it for the client, deploy it, watch it in production, and fix what breaks. The FDE model means you own outcomes, not tasks.
Requirements:
The tech you will work with:
- Frontend: React, React Flow (node-based graph editors).
- Backend: Python, Node.js, FastAPI.
- Data: ClickHouse, Kafka (event streaming), Debezium (CDC).
- AI/ML: LangGraph (agent orchestration), LLMs (prompt engineering, RAG, tool use), DPT-2 (document extraction), PageIndex (long-document reasoning), Weaviate (vector search).
- Permissions: OpenFGA (Zanzibar-based relationship access control).
- Observability: Langfuse (LLM tracing), OpenTelemetry.
- Cost control: LiteLLM (gateway + budgets), Orkestra (model routing).
- Infrastructure: dbt (semantic layer / MetricFlow), Temporal (scheduling), DataHub / OpenMetadata (catalogue).