Cloud & AI Engineer · London, UK
Raian Khan
IT Engineer at Odgers · MSc AI, Liverpool
Building production AI systems and cloud-native infrastructure. Fine-tuned LLMs, RAG pipelines, and AWS deployments from concept through to live.
IT Engineer at Odgers · MSc AI, Liverpool
Building production AI systems and cloud-native infrastructure. Fine-tuned LLMs, RAG pipelines, and AWS deployments from concept through to live.
About
I'm an IT Engineer at Odgers in London, building production AI systems alongside enterprise infrastructure work: fine-tuned LLMs, RAG pipelines, full-stack AI applications, and cloud-native ML inference platforms on AWS.
Day to day I manage a hybrid AWS, Azure, and M365 environment spanning EMEA, APAC, and AMER, and I've shipped an LLM-based triage system internally that drove a roughly 35% improvement in ticket handling efficiency.
Outside of work I'm building two startups: FinLit, a bank-agnostic AI personal finance assistant, and Gridient, a scheduling optimisation platform for power-constrained data centres.
Currently pursuing an MSc in Artificial Intelligence at the University of Liverpool and working towards AWS Machine Learning Associate certification.
Projects
Production AI systems, ML infrastructure, and full-stack applications built end-to-end.
Startup building a bank-agnostic AI personal finance assistant. Ingests transaction data from any UK bank CSV export with auto-categorisation across 20+ spending categories. Privacy-first architecture: raw transactions processed locally, only aggregated summaries sent to Claude Haiku for insight generation (approx. £0.11/user/month). Includes a compound growth simulator modelling opportunity cost of spending over 1 to 30 year horizons. Built with Next.js 14, FastAPI, PostgreSQL, and Docker.
Scheduling optimisation platform for data centres in high-sun, power-constrained markets. Pulls solar forecasts, grid carbon intensity, battery state of charge, and electricity pricing, then uses an OR-Tools CP-SAT solver to decide when to run compute workloads so operators burn less diesel and cut emissions. Covers five facilities across Lagos, Nairobi, Dubai, Mumbai, and San Francisco, with proxy models filling the gap where nobody else signals well (fixed diesel intensity for Lagos, rainfall-modulated hydro for Nairobi). XGBoost for solar correction, React dashboard on Vercel.
Fine-tuned Qwen2.5-0.5B-Instruct using LoRA adapters on ~800 synthetic samples for structured IT ticket classification. Deployed as a FastAPI microservice achieving ~95% schema-valid JSON output compliance. Born from a real operational bottleneck at Odgers, the production version drove a ~35% improvement in ticket handling efficiency.
A retrieval-augmented generation backend built on 384-dim SentenceTransformer embeddings, IVFFlat-indexed pgvector store, and cosine similarity top-k retrieval hitting sub-100ms latency. Fully documented REST API with typed request/response contracts, ready to be dropped behind any LLM front-end.
Production ML inference API on AWS ECS Fargate behind an ALB, with full infrastructure-as-code via Terraform and a CI/CD pipeline through GitHub Actions using AWS OIDC keyless auth. Designed for zero-downtime deployments with automatic container image promotion through ECR.
Skills
The tools and technologies I work with.
AI / ML
Backend & Full-Stack
Cloud & Identity
Currently studying
Contact
Open to new opportunities, collaborations, and interesting conversations. Drop me a line.
hello@raian.uk