Remote job
AI Engineer, LLM Applications
Job details
About this role
Role overview
Develop and ship language model features as part of product teams, with responsibility for retrieval quality, evaluation, monitoring, and safe failure handling. The work includes conversational assistants and agent workflows, while applying judgment about when conventional search or better data is a more suitable solution.
Responsibilities
- Design retrieval systems using techniques such as chunking, embeddings, hybrid search, and reranking. - Build agent workflows with constrained tool access, clear boundaries, and human handoff where needed. - Create evaluation suites that track answer quality, cost, and latency across releases. - Add safeguards against prompt injection, data leakage, and unsupported or risky responses. - Version prompts and connect them to evaluations and deployable application code. - Diagnose model behavior in production and recommend simpler approaches when they better serve users.
Requirements
- At least two years shipping production software in Python or TypeScript, including an LLM feature used by real users. - Practical retrieval experience with embeddings, hybrid search, and reranking, using a vector store such as pgvector or Qdrant. - Ability to build structured outputs and tool calls with schemas and validation. - Production-quality coding practices: typed, tested, packaged code that other engineers can deploy and maintain. - Experience with model providers or open-weight models, prompt versioning, and evaluation practices. - Understanding of guardrails and failure modes, including injection, leakage, refusals, and fallback behavior.
Nice to have
- Fine-tuning, distillation, or production operation of open-weight models. - Speech, document extraction, or other multimodal applications. - Familiarity with regional data residency requirements.
Benefits and work setup
- Remote work across the MENA region, with hours arranged around team overlap. - Small teams with direct client contact and products that typically ship in weeks. - Budget for equipment and tools, plus a learning budget with time to use it. - Paid time off and no timesheet-based tracking.