Senior Backend (NodeJS) Engineer

This role is suitable for a Senior Backend Engineer with 8+ years of commercial Node.js experience who can design, scale, and own complex production backend systems.

The candidate should have strong skills in backend architecture, APIs, PostgreSQL, AWS, observability, testing, CI/CD, and integrating production-grade LLM, multi-agent, and RAG solutions with reliable, cost-efficient infrastructure.

They should be familiar with designing scalable services, AI-driven automation workflows, agent orchestration, retrieval and ranking pipelines, prompt evaluation, monitoring, debugging, and cross-functional product development, ideally in healthcare or regulated environments.

How would you design a scalable Node.js backend architecture for AI-powered clinical automation workflows?

I would decompose the system into independently deployable services or modules around workflow orchestration, AI integrations, data access, and external APIs. I would use asynchronous processing where appropriate, define clear contracts, apply idempotency and retries, and scale stateless components horizontally while keeping clinical data consistency and auditability in mind.

How would you make an LLM integration reliable enough for production use?

I would enforce structured outputs with JSON Schema validation, implement timeouts, retries with backoff, fallbacks, and circuit breakers, and treat model responses as untrusted input. I would also add tracing, evaluation datasets, quality metrics, prompt and model versioning, and cost and latency monitoring.

When would you use a multi-agent architecture instead of a single LLM workflow?

A multi-agent design is justified when the workflow contains clearly separable responsibilities, different tools or policies, and decision-making that benefits from specialized agents. I would avoid it for simple flows because additional agents increase latency, cost, coordination complexity, and failure modes.

How would you implement a supervisor-agent pattern in Node.js?

The supervisor would maintain workflow state, select specialized agents, validate their outputs, and decide whether to continue, retry, escalate, or terminate. Each agent would expose a typed contract and controlled tool access, while the orchestrator would enforce timeouts, iteration limits, permissions, and audit logging.

How would you design and optimize a RAG pipeline?

I would evaluate document chunking, metadata extraction, embedding models, vector indexing, filters, hybrid retrieval, reranking, and context assembly as separate stages. Optimization should be driven by retrieval and answer-quality evaluations while balancing relevance, token usage, latency, and infrastructure cost.

How would you prevent hallucinations and unsafe actions in an AI healthcare workflow?

I would constrain responses to validated schemas and trusted context, require evidence or citations where applicable, and use deterministic business rules for safety-critical decisions. High-risk actions should require explicit authorization or human review, with complete audit trails and monitoring for policy violations.

How would you build an AWS-based backend using Lambda, API Gateway, ElastiCache, and PostgreSQL?

API Gateway would expose authenticated APIs, Lambda would handle stateless request and event processing, PostgreSQL would store transactional data, and ElastiCache could support caching, rate limiting, or short-lived workflow state. I would design for connection pooling, idempotent handlers, asynchronous queues where needed, least-privilege IAM, and clear operational limits.

What PostgreSQL techniques would you use to maintain performance at scale?

I would design appropriate indexes and constraints, inspect query plans with EXPLAIN ANALYZE, avoid inefficient joins and unbounded queries, and use pagination and batching. I would also manage migrations safely, monitor slow queries and connection usage, and consider partitioning or read replicas when justified by workload characteristics.

How would you implement observability for an AI-driven backend?

I would collect structured logs, metrics, and distributed traces across API calls, workflow steps, model requests, retrieval operations, and external integrations. Grafana and OpenSearch could support dashboards and alerting for errors, latency, token usage, cost, retrieval quality, failed agent steps, and unusual clinical workflow behavior.

How would you test and continuously evaluate an LLM-based feature?

I would combine unit and integration tests for deterministic backend logic with contract tests for tools and schemas, replay tests for representative conversations, and regression evaluation sets for model behavior. Quality, safety, latency, and cost metrics should be tracked across prompt or model changes before and after deployment.

MEDvidi is an AI-powered mental healthcare clinic building safe, scalable telemedicine solutions for patients and providers across the United States. As a Senior Backend (Node.js) Engineer, you will design and scale AI-driven integrations, multi-agent workflows, RAG pipelines, APIs, and reliable clinical automation systems. The stack includes Node.js, PostgreSQL, AWS CDK, Lambda, API Gateway, ElastiCache, Grafana, OpenSearch, and modern LLM tooling. This is a fully remote, full-time EU role with a competitive salary based on experience, offering high impact, strong professional growth, wellness benefits, and the opportunity to shape the future of healthcare AI.