About The Role
The role focuses on building production-grade LLM and generative AI systems - RAG pipelines, agentic workflows, fine-tuning infrastructure, and evaluation frameworks that ship to real users. This is hands-on engineering, not prompt tinkering: the work spans retrieval architecture, model optimization, and serving infrastructure where latency, cost, and reliability all matter.
The position sits on a small, senior AI team with direct influence over architecture decisions and a clear path from prototype to production. Models and pipelines built here power customer-facing product features at scale.
Key Responsibilities
- Design and ship production RAG systems - retrieval strategies, chunking, reranking, hybrid search - using LangChain, LlamaIndex, or custom architectures
- Build and tune vector database integrations (Pinecone, Weaviate, pgvector) and embedding pipelines for semantic search at scale
- Develop agentic workflows and tool-use pipelines with robust error handling, state management, and guardrails
- Implement systematic LLM evaluation: benchmark suites, LLM-as-judge pipelines, regression testing, and hallucination detection
- Run parameter-efficient fine-tuning (LoRA, QLoRA) and instruction tuning on domain-specific datasets; manage the full training-to-deployment loop
- Optimize inference cost and latency through quantization, batching, caching, and model routing across open-weight and frontier APIs
- Monitor production LLM systems for drift, cost anomalies, and quality regressions; contribute to team engineering standards and architecture reviews
What We Are Looking For
- 3-8 years of software engineering or ML engineering experience, including at least 2 years working with LLMs in production
- Strong Python skills; comfort with async programming, API design, and cloud infrastructure (AWS, GCP, or Azure)
- Hands-on experience with RAG architectures, vector databases, and embedding models in production environments
- Practical experience fine-tuning or adapting open-weight models (Llama, Mistral, Qwen) using PEFT methods
- Familiarity with LLM observability and evaluation tooling (LangSmith, Langfuse, Ragas, or homegrown equivalents)
- Bachelor's degree in Computer Science, Engineering, or equivalent practical experience; track record of shipping AI features end-to-end
- Bonus: Experience with Kubernetes-based model serving (vLLM, TGI, Triton), distributed training with PyTorch, or publishing/applied research in NLP
