Identity
I design and ship end-to-end ML systems under real constraints — from data pipelines and statistical modelling to CI/CD for ML, inference serving, and monitoring-ready deployment. Where the use case demands it, I build RAG/LLMOps workflows with retrieval tracing and evaluation harnesses.
I treat ML as a product system, not a notebook. I ship with reproducibility, measurement, and operational clarity: data validation, versioned artefacts, and deployment discipline. I’m deliberate about trade-offs — latency, cost, robustness, and auditability — and I prefer evidence over claims.
Ingestion → data modelling → feature engineering → training + evaluation → artefact registry → containerisation → serving APIs → monitoring and retraining triggers. For GenAI use cases: retrieval, chunking, reranking, guardrails, and evaluation harnesses.
This site highlights systems that connect modelling quality with engineering quality. The goal is to demonstrate a single identity: a full-stack ML engineer who can both build models and deploy them reliably.
Supervised learning, feature engineering, uncertainty-aware thinking, model evaluation, calibration, and robust baselines. When using LLMs: retrieval-augmented generation, prompt + tool orchestration, and evaluation-driven iteration.
CI/CD for ML, container-first delivery, reproducible training, deployment automation, and monitoring-ready outputs. I design for reliability: versioning, provenance, and fast rollback paths.
Full Stack Machine Learning Engineer / Data Scientist building scalable ML pipelines and deployment systems. Keywords: CI/CD for ML, Dockerisation, model serving, inference scaling, monitoring, automated retraining, RAG/LLMOps, evaluation harness, data governance, and production trade-offs.
