Aleph Staff+ Software Engineer at Aleph owning AI foundation systems including model selection, routing, and evaluation infrastructure. The role involves shipping production LLM systems and building agentic features end to end.
Responsibilities
Own the AI foundation product teams build on: model selection and routing, the model proxy layer, context management, and tool design
Own eval infrastructure. Work with customers and finance domain experts to define what excellent output looks like, then encode it in evals every team can run - in FP&A, a wrong answer ends up in someone's board deck
Ship agentic features end to end, from v0 through the optimization loop: prompt and context engineering, caching, parallel tool calls, subagent patterns
Build observability into agent behavior so we can profile it, find bottlenecks, and decide what to fix with data
Track what's changing in agentic systems and bring the practices that prove out into how we build
Qualification
You build proof of conceptsStrong judgment about abstractionsYou want to ship production systemsAdoptionStrong fundamentalsHigh agencyPragmatismCraftCommunication
Required
You've shipped production LLM systems - evals, context management, multi-model or multi-provider routing - and can talk concretely about what broke and what you changed
You build proof of concepts, decide quickly, and ship v0s. Evals drive your quality bar
Strong judgment about abstractions: opinionated about design, pragmatic about shipping incrementally
Experience with LLM APIs and agent frameworks, and shipped user-facing products
You want to ship production systems, not do research
What we expect from all Staff+ Engineers
Adoption: teams you don't manage have picked up systems you built - because they were that good
Strong fundamentals: you can trace requests from UI to data store and make good technical decisions at every layer
High agency: background as a founder, engineering lead, or startup builder. You have a builder's mindset and find creative ways to ship
Pragmatism: you choose the fewest and simplest tools that will actually work
Craft: you care deeply about code quality and write code that's correct and easy to understand
Communication: clear, direct, and persuasive across technical and non-technical audiences
