About the Role:

We are looking for a motivated AI/ML Engineer to work across LLM and SLM training, visual defect detection using YOLO models, and multi-agent system development. You will own the full lifecycle from data preparation and model training to production deployment, with a focus on building reliable, efficient, and domain-specific AI systems.

Key Responsibilities:

LLM & SLM Training:

- Fine-tune large and small language models (LLMs & SLMs) on domain-specific datasets using SFT, LoRA, and QLoRA.

- Train lightweight SLMs (1B7B parameters) for specific use cases such as defect classification, report generation, anomaly summarisation, and structured data extraction.

- Apply preference alignment techniques RLHF and DPO to align model outputs with task requirements.

- Build and maintain RAG pipelines to ground model responses in domain knowledge.

- Evaluate models against task-specific benchmarks and iterate on training data quality.

- Optimise inference using vLLM, TGI, or llama.cpp for cost-efficient production serving.

Visual Defect Detection YOLO Models:

- Train and fine-tune YOLO models (YOLOv8, YOLO11, YOLO26) for defect detection, segmentation, and classification.

- Build annotation pipelines and manage image datasets using Roboflow or Label Studio.

- Optimise models for edge and CPU deployment using TensorRT, ONNX, or OpenVINO.

- Develop monitoring and retraining workflows to handle real-world data drift in production.

Multi-Agent System Development:

- Design and build multi-agent workflows using LangGraph, CrewAI, or AutoGen.

- Define agent roles, implement tool use and function calling, and manage state across agent turns.

- Integrate agents with external APIs, databases, and internal services.

- Build evaluation and oversight mechanisms for agent reliability and safety in production.

MLOps & Deployment:

- Package and deploy models as REST APIs using FastAPI, containerised with Docker.

- Track experiments and model versions with MLflow or Weights & Biases.

- Set up cloud-based training and serving pipelines on AWS, GCP, or Azure.

- Maintain documentation model cards, data sheets, and experiment logs.