Senior AI Engineer for LiteLLM Gateway + Custom SDK (Fine-Tuning & Evals)

We’re looking for an AI engineer who can join an existing codebase and workflow — not someone who wants to stand up a new agent stack from scratch. You will get a Devin account and use it as the main way you ship code. The immediate work is on two production systems (although this could grow substantially): 1. A custom agent SDK we are building from scratch (not a LangChain, LlamaIndex, or similar wrapper) 2. Our LiteLLM gateway, including routing, fallbacks, budgets, API keys, and logging Must-haves — please link evidence in your proposal, not just list tools: - You have fine-tuned an open-weight LLM yourself (LoRA, QLoRA, supervised fine-tuning, and ideally DPO). Tell us the base model, dataset size, and before/after metric. - You have built a custom eval harness: fixed test sets, a calibrated LLM-as-judge or equivalent, and CI checks that catch regressions when a prompt, model, or code path changes. - You have designed and published an SDK or library on PyPI or npm, with versioning, typed interfaces, retries, auth, and docs. “I built agents with LangChain” does not count. - You have run LiteLLM or a similar OpenAI-compatible gateway in production. If you haven’t used LiteLLM itself, tell us what you ran instead and how you handled routing, fallbacks, spend, keys, and logging. - You are comfortable working in someone else’s repository through written specs and pull requests. Nice-to-haves: Devin, Claude Code, Cursor, or Codex in a spec → review → pull request loop; vLLM; tracing tools such as Langfuse; TypeScript as well as Python; working hours that overlap UTC-4. Please answer the screening questions with links and concrete detail, not just a list of tools. This is not a chatbot, RAG-demo, or no-code automation job. To be clear - we are looking for a true 10x engineer. We have an active dev capability in-house and your OUTPUT will be benchmarked against what we can do internally. For the successful hire this could turn into alot of ongoing work. 5-10 hrs a week to start.