Backend AI Engineer
Posted Updated
Description
Responsibilities:
- Ingest and process high-volume, multi-source data across enterprise systems.
- Validate stateful business workflows spanning loosely coupled services.
- Reconcile data across sources with consistency and correctness guarantees.
- Build event-driven, asynchronous pipelines for real-time and batch validation.
- Integrate with complex external systems (e. g., SAP, APIs, data warehouses).
You'll also work on:
- Deploying LLM-powered agents into production workflows.
- Designing systems to evaluate, monitor, and improve agent behavior over time.
- Building safeguards for correctness, reliability, and failure handling.
- Ensuring non-deterministic AI outputs meet deterministic system requirements.
- You'll own problems end-to-end: design → build → operate.
Tech (evolving):
- Languages: Python, Go, Node.js .
- Data: Snowflake, Databricks, Postgres.
- Architecture: distributed systems, event-driven pipelines.
- Infra: containers + cloud platforms (Kubernetes / ECS / Cloud Run).
- AI: LLM-based agents with evaluation, monitoring, and control systems.
Requirements:
- Production builder: You've shipped LLM-powered features real users depend on and debugged them when they broke.
- LLM practitioner: You understand hallucinations, retrieval failures, context limits, and what it takes to make agents deterministic enough for enterprise use.
- Systems thinker: You design for latency, failure modes, retry logic, and observability before features.
- Enterprise-aware: Data residency, compliance, audit trails, and deterministic guardrails are first-class design constraints for you.
- 3+ years in AI/ML or backend engineering with strong AI exposure.
- Hands-on production experience with LLM APIs (Anthropic, OpenAI, Cohere).
- Experience designing evaluation frameworks: automated evals, regression tests, or LLM-as-judge pipelines.
- Strong Python; experience with LangChain, LlamaIndex, or similar agentic frameworks.
- Familiarity with RAG architectures: chunking, embedding models, vector DBs, retrieval quality.