Senior Software Engineer, AI Platform

At Mirai, a Scopely Studio, we're passionate about creating engaging gaming experiences that delight millions of players worldwide. We are looking for a Senior Software Engineer, AI platform, who builds products and can see the platform inside them. You will write product features yourself, and you will turn what repeats across products into shared services and libraries the rest of the team builds on. The platform comes out of real product work, one product at a time, not ahead of it.

Our AI Native Engineering team builds software for Scopely game teams and internal departments: tools that use large language models (LLMs) for bug triage, player feedback analysis, market and company research, and contract analytics, alongside direct-to-consumer web stores for Scopely games. Most of these tools repeat one pattern: connect a source system, build a trusted store, put an LLM or governed logic over it, and control who can see what. Today each product builds that pattern its own way, with its own model setup, its own access rules, and its own safety controls.

This is a software engineering role, not an infrastructure role. Our DevOps and SRE engineers own clusters, pipelines, and cloud provisioning. If your recent work is mainly infrastructure with little product code, this is not the right fit.

What you will do

• Build product features end to end on our internal AI tools, in TypeScript or Python, from design through production.
• Find what our products build separately and turn it into shared services and libraries: model access and spend, identity and permissions, retrieval, tool servers, and the controls around LLM calls. Move products onto them one at a time, without stopping delivery.
• Build MCP tool servers that expose company data to agents and chat tools, with the end user's permissions enforced before any data reaches the model.
• Help introduce relationship-based access control across products, so access rules are defined once and applied to both database queries and vector search.
• Standardize model access through our LLM gateway, so model choice, cost, and usage are set in one place rather than in each repository.
• Spread the safety patterns that live in one project each today to every product: prompt-injection handling, kill switches, sandboxed execution of generated code, and locked-down data access.
• Make the shared pieces easy to adopt: clear APIs, versioning, templates, and documentation another engineer can follow without asking you.
• Set the standard in code review, including for AI-assisted code, and mentor engineers through review, pairing, and design.
• Own incidents on what you build, and work with our DevOps engineers on how it is deployed and run.
What we are looking for

• Six or more years building and running production software, including at least one shared service, library, or internal API that other engineers or teams depended on.
• Strong TypeScript and Node.js, and strong Python. Our services are split across both.
• Shipped LLM features to production: retrieval over embeddings, tool calling, structured output, and a way of checking output quality.
• Designed APIs or libraries other engineers build on, with real attention to versioning, compatibility, safe defaults, and documentation.
• Built multi-tenant or permission-sensitive systems, and knows where to enforce access so data never reaches a user, or a model, that should not see it. Treats model output and retrieved content as untrusted input.
• Solid system design: service boundaries, data modeling, failure handling, idempotency, and cost.
• Production experience on AWS, with infrastructure as code in Terraform or similar, enough to own how your services ship alongside DevOps.
• Writes tests as part of the work and treats CI checks as gates, not suggestions.
• Clear written and spoken English, and able to explain technical tradeoffs to product and business stakeholders.
• Uses AI coding tools and verifies their output. You understand, can prove, and own what you merge.

The Stack

Languages: TypeScript and Node.js, Python 3.11.

Services: Express, FastAPI, Next.js

AI: OpenAI models and embeddings, LangChain, access control with SpiceDB (designed, being introduced)

Model Context Protocol (MCP) SDK, an internal LLM gateway built on LiteLLM over Amazon Bedrock

Cloud and Delivery: AWS (EC2, ECS Fargate, Lambda,

Data: PostgreSQL (Amazon RDS and Aurora), BigQuery, Amazon OpenSearch Serverless for vector search

App Runner, S3, CloudFront), Terraform, GitLab and GitHub

Identity and Access: Okta, OAuth 2.0 with PKCE, relationship-based access control with SpiceDB (designed, being introduced)

Work Tracking: Jira, Confluence

Nice to Have

• Relationship-based access control in practice, with SpiceDB, OpenFGA, or similar
• An LLM gateway such as LiteLLM or Amazon Bedrock, and evaluation or tracing tools such as Braintrust, Langfuse, or LangSmith
• An agent orchestration framework such as LangGraph
• A shared platform that served several teams or departments rather than one product
• Experience in games, or Arabic

• Mobile Phone Allowance
• Health Insurance
• Performance Bonus
• Relocation support where applicable