I Speak About AI & Data

How do you go from a proof-of-concept chatbot to an enterprise-grade retrieval-augmented generation system? This talk walks through designing vectorisation pipelines, choosing retrieval strategies (dense, sparse, hybrid), and orchestrating multiple specialised agents with LangChain and LlamaIndex. I cover the hard lessons learned shipping multi-agent RAG at scale — from chunking trade-offs and embedding model selection to guardrails, evaluation frameworks, and the operational patterns that keep production AI reliable.

Most ML teams start with notebooks and good intentions. This talk is about what comes next: standing up production ML infrastructure on Kubernetes with Kubeflow, orchestrating training and feature pipelines through Airflow, versioning data and models with DVC, and the YAML DAG factory pattern that lets data scientists ship pipelines without writing boilerplate code. I share the architecture decisions, tooling trade-offs, and cultural shifts that turned a scrappy data team into a self-service ML platform.

My career started in quality assurance, moved through test automation and back-end engineering, took a detour into computer vision, and landed at leading AI teams. This talk is about the power of non-linear career paths — how switching domains forces you to learn faster, why breadth of experience makes you a better technical leader, and how to navigate the uncertainty of reinventing yourself every few years. Practical advice for engineers who feel stuck in a single lane.

The AJNA system stitches two 8K camera feeds into a single panoramic video and uses computer vision to track players, measure ball speed, and analyse team formations — all in real time, without human operators. This talk covers the end-to-end pipeline: camera calibration and image stitching, object detection and tracking models, the engineering challenges of processing 8K video at broadcast latency, and how we built a system that turns raw footage into structured sports analytics.

AI teams are not typical software teams. They blend ML engineers, software developers, data engineers, and DevOps specialists — each with different workflows, tools, and definitions of “done”. This talk shares what I have learned about hiring for cognitive diversity, designing interview loops that actually predict on-the-job performance, mentoring engineers across disciplines, and creating a team culture where research rigour and shipping velocity coexist. Concrete frameworks for managers building or scaling AI organisations.