Fluvius

Project Overview

We are building a production-grade AI Operations Platform — real infrastructure, not API wrappers. We need a high-caliber Full-Stack Engineer with an ownership mindset and a passion for clean architecture to join a small, focused team with end-to-end responsibility.

This is a hands-on role. You will architect, build, and ship complex AI agent pipelines that serve real clients at scale. If your AI experience is limited to internal tools, hobby projects, or basic prompt-and-response integrations — this is not the right fit. We need someone who has already navigated the pain of production-scale AI systems and knows how to solve the hard problems that come with it.

Technical Stack

• Backend (Primary): Expert-level Node.js — NestJS preferred, but strong Express or Koa experience is equally welcome. You will own the entire server-side architecture.
• Database: MongoDB — advanced indexing, data modeling, aggregation pipelines, multi-tenant isolation.
• Frontend: Next.js / React. Ability to build responsive, production-quality UIs.
• AI / LLM Layer: Production AI agent development: complex multi-step pipelines, tool-calling agents, streaming LLM responses, context management, agentic loops with safeguards.
• Message Queues: Redis, RabbitMQ, or equivalent — for decoupling LLM streaming, buffering, and async processing.
• Infra / DevOps: Solid understanding of cloud deployments, CI/CD, containerization (Docker), and basic monitoring.

Key Accountabilities

As a Senior Full-Stack Developer on this platform, you will be directly accountable for:

• AI Agent Architecture & Reliability: Design, build, and maintain complex, multi-step AI agent pipelines that serve production traffic. Implement robust safeguards against infinite agentic loops (iteration caps, token budgets, timeout constraints). Ensure agents behave predictably under load and edge cases.
• LLM Streaming & Data Persistence: Implement real-time LLM response streaming to end users via WebSockets or SSE without blocking the UI. Architect a reliable server-side data pipeline: buffer streaming chunks through Redis or a message queue, then persist complete responses to MongoDB. Guarantee zero data loss even under high concurrency.
• Multi-Tenant Data Isolation & Security: Design and enforce strict tenant-based data isolation so that AI agents can never leak or cross-reference data between clients. Implement tenant scoping at the database query level (not just prompt-level instructions to the LLM). Own security architecture for sensitive client data flowing through AI pipelines.
• End-to-End Feature Ownership: Take features from requirements through architecture, implementation, testing, and deployment. Write clean, maintainable code with proper separation of concerns. Review your own work critically before it ships.
• Backend & API Design: Build and maintain well-structured Node.js services (NestJS, Express, or Koa) with clean module separation, guards/middleware, and proper layering. Design efficient MongoDB schemas, indexes, and aggregation pipelines for high-throughput AI workloads.
• Frontend Delivery: Develop responsive, performant Next.js / React interfaces that surface AI agent interactions to end users. Handle real-time data display (streaming responses, status updates, progress indicators).
• Infrastructure & DevOps Participation: Manage deployments, CI/CD pipelines, and containerized environments. Monitor system health, set up alerting, and participate in incident response.
• Communication & Collaboration: Communicate clearly and proactively with stakeholders. Raise blockers early. Participate in architecture discussions with a thoughtful, structured approach.

Requirements

Must-Have (Non-Negotiable):

• At least 4 years of professional full-stack development experience.
• Expert-level Node.js — NestJS preferred, but strong experience with Express or Koa is equally accepted. You can architect a complex backend from scratch.
• Strong MongoDB skills — data modeling, indexing strategies, aggregation framework, multi-tenant patterns.
• Production experience with AI agents — not just calling an API with a prompt. You have built and shipped multi-step agentic workflows that serve real, external users at scale. You understand agentic loops, tool-calling patterns, context windowing, and the failure modes that come with production AI.
• Hands-on experience with LLM streaming — you have implemented real-time token streaming to users and solved the server-side persistence problem (buffering via Redis/queues → MongoDB).
• Multi-tenant data isolation — you have implemented tenant-scoped data access in a system where AI agents interact with sensitive, per-client data. You understand why prompt-level isolation is insufficient.
• Next.js / React proficiency — you can build and ship production frontends.
• Clean architecture mindset — separation of concerns, modular design, readable code.

Strong Nice-to-Have:

• Experience with Redis, RabbitMQ, or similar for async processing and event-driven architectures.
• Familiarity with vector databases (Pinecone, Weaviate, pgvector) for RAG pipelines.
• Experience with LangChain, LangGraph, CrewAI, or similar agent orchestration frameworks.
• Proficiency with AI-assisted development tools (Cursor, GitHub Copilot) for velocity.
• Understanding of cost management for LLM API calls at scale (token budgets, caching strategies, model routing).
• Cloud infrastructure experience (AWS, GCP, or Azure) including containerization and orchestration.
• Monitoring and observability for AI systems (tracing agent decisions, logging tool calls, measuring latency).

Who You Are

• Battle-Tested: You have already dealt with production AI agent failures — infinite loops, data leaks, streaming interruptions, cost explosions — and you know how to prevent them. We do not have time to train on these problems.
• Architecturally Minded: You think in systems, not just features. You design for scale, maintainability, and failure recovery from day one.
• Professional & Patient: You are a calm, inclusive communicator who handles complex problems with patience and clarity. You can explain technical decisions to non-technical stakeholders.
• High Work Ethic: Reliable, independent, and committed to high-quality delivery. You do not need to be micromanaged.
• Ownership-Driven: You treat the product as your own. You proactively identify risks, suggest improvements, and take initiative.

What to Expect in the Interview

We will assess your production AI experience through scenario-based technical questions. Be prepared to discuss:

• How you prevent AI agents from entering infinite execution loops and the specific safeguards you implement (iteration limits, token budgets, timeout mechanisms).
• Your approach to streaming LLM responses to users in real time while ensuring complete, lossless persistence on the server side. We expect you to articulate the full pipeline: LLM → buffer (Redis/queue) → database → UI.
• How you enforce data isolation between tenants when AI agents access per-client data. We expect you to explain tenant-scoped database access, not just prompt instructions.
• Real production incidents or challenges you faced with AI agents at scale and how you resolved them.
• Your architectural decisions on a past project: why you chose specific patterns, what tradeoffs you made, and what you would do differently.

This is not a quiz. We want to see evidence that you have operated in this space and can navigate its unique challenges with confidence.