About RiverX
RiverX is an AI platform for building and evolving websites and web applications. It brings model selection, coding agents, repository-backed projects, live previews, database provisioning, and publishing into a connected workflow. People should be able to move from an idea to a working application, understand what the agent changed, and keep improving it without losing control of their code.
Our product spans a Next.js and React frontend, a Node.js backend, agent and workspace services, Supabase for authentication and core application data, and Turso for project databases and the skills catalog. GitHub, Railway, Vercel, and external model providers are part of the engineering environment. Each role below owns a specific part of making that experience dependable.
The role
You will own the request path between RiverX’s agent workflows and external model providers. Today, RiverX integrates provider APIs and model catalogs; this opening is centered on inference integration and systems reliability. You will make model requests behave predictably across streaming, tool calls, cancellation, provider failures, and usage reporting.
What you will own
- Build and improve provider adapters and request handling for model-backed coding workflows, preserving model-specific capabilities and response semantics.
- Make streaming robust across slow consumers, disconnects, partial responses, timeouts, cancellation, and provider errors. Prevent abandoned requests from continuing unnecessarily.
- Design and evaluate retry and fallback policies that respect rate limits, request compatibility, tool-call state, and the risk of duplicated work.
- Connect model and provider metadata to request-time decisions with explicit behavior for missing, stale, or incomplete data. Validate routing changes before expanding their rollout.
- Instrument time to first token, end-to-end latency, throughput, request success, task success, token usage, and cost. Keep measurements and provider-reported estimates distinguishable.
- Preserve the provider, model, credential source, and usage associated with an attempt, including bring-your-own-key flows, so billing and debugging can rely on accurate attribution.
- Improve credential isolation, request authorization, observability, and safe diagnostics without exposing provider secrets or user content unnecessarily.
- Work with applied AI engineers to judge inference changes by useful task completion, not a single latency or price metric.
What you should bring
- Strong experience with networked services, API gateways, inference integrations, or another latency-sensitive request path.
- Proficiency in TypeScript/JavaScript, Python, Go, or a comparable systems language, with willingness to contribute to a Node.js platform.
- A detailed understanding of HTTP streaming, timeouts, backpressure, connection lifecycles, retries, and concurrency control.
- Experience integrating model APIs, including tool calling, context limits, response formats, token accounting, or provider-specific behavior.
- The ability to design reproducible load and failure tests and interpret latency distributions rather than relying on averages alone.
- Careful judgment about correctness when a request partly succeeds or a provider omits usage information.
Useful additional experience
OpenAI-compatible APIs, OpenRouter, distributed tracing, usage-based billing, load testing, or GPU inference systems. Experience with batching, KV caches, quantization, or model serving is useful, but operating a self-hosted GPU fleet is not a prerequisite for this role.
Early priorities
Trace a model request from agent initiation through streaming and usage settlement. Establish a reproducible baseline, address one important failure mode, and propose a routing or efficiency improvement with explicit fallback behavior and measurable acceptance criteria.
Include in your application
An inference, gateway, or high-throughput service you worked on. Describe a performance tradeoff and how you protected correctness while improving it.
Working together
This is a full-time, remote role. We work through clear written proposals, reproducible examples, thoughtful code or design reviews, and direct collaboration when a problem needs it. You should be comfortable owning an outcome across implementation, validation, documentation, and follow-through. Please include your location and time zone in your application so we can discuss practical collaboration arrangements.
How to apply
Apply through this page with your résumé as a PDF. Add links to relevant work and a short note explaining your contribution to a project, a difficult decision you made, and what you learned. Work can come from employment, open source, independent projects, or research. Share only material you have permission to share; please do not include customer data, credentials, or confidential code.