About the role
Responsibilities:
- Design, build, and ship end-to-end LLM-powered features — RAG, agentic workflows, tool calling, and multi-step AI pipelines — using LangChain and LangGraph.
- Build and maintain production-grade AI services and APIs in Python and FastAPI.
- Design and optimise retrieval systems — document ingestion and chunking, embeddings, vector databases, hybrid search, and reranking — that hold up at production scale.
- Work across commercial LLM APIs (OpenAI, Anthropic, Gemini) and open-source models (such as Llama, Qwen, and Mistral), selecting the right model for each task based on accuracy, latency, cost, and data-privacy needs.
- Fine-tune open-source LLMs and embedding models on domain data — from dataset preparation and training to evaluation against baselines.
- Deploy and serve models on GPU infrastructure — containerised with Docker, orchestrated on Kubernetes, and served through vLLM or equivalent — optimising for throughput, latency, and cost.
- Establish evaluation and monitoring for AI features: define quality metrics, build evaluation sets, diagnose failure modes, and drive continuous improvements in accuracy and reliability. An Intelligent layer for global trade runs on.
Requirements:
- Bachelor's degree in Computer Science, Engineering, or a related quantitative discipline.
- 4-5 years of experience building AI/ML systems, with at least 2 years building and shipping LLM-powered applications in production.
- Strong proficiency in Python, with hands-on experience building and containerizing production APIs using FastAPI and Docker.
- Demonstrated production experience with LangChain and LangGraph, RAG, agentic workflows, and tool calling.
- Hands-on experience with vector databases (such as pgvector, Qdrant, etc. ) and production retrieval pipelines.
- Practical experience with open-source LLMs alongside commercial APIs, including fine-tuning with LoRA, QLoRA, or PEFT using PyTorch and Hugging Face Transformers.
- Experience serving LLMs on GPU machines with vLLM, TGI, or equivalent, including quantisation, on Kubernetes and at least one major cloud platform (AWS, Azure, or GCP).
- Experience with MLOps/LLMOps practices — experiment tracking and model versioning (such as MLflow or Weights and Biases), CI/CD for models and prompts, and safe rollout and rollback of model updates.
- Experience with document AI — OCR, layout parsing, or data extraction from scanned documents.
- Experience with evaluation and observability tooling for production AI systems (such as LangSmith, Langfuse, or RAGAS).
- Clear communicator who works autonomously in a fast-paced startup and can separate substance from hype.