Muhammad Azam
AI Enablement Lead & Technical AI Lead
I build AI systems that make it into production and help the engineering teams around me do the same.
About
I've spent the last ten years in backend engineering, and the past couple getting deep into AI. I build RAG systems, LLM agents and semantic search platforms for UK government and enterprise clients. More recently I've been figuring out how to help organisations actually adopt this stuff in a way that sticks.
Right now I'm embedded across government engineering teams at DSIT. I go in for a few weeks at a time, helping them get RAG pipelines running, sorting out their AI delivery process and setting up tools like Claude Code and GitHub Copilot. I lead a team of six and we've got systems in production handling around 4,000 queries an hour. The playbooks and guides I put together tend to get picked up by other teams too, which is the whole point.
My background is Java, Spring Boot and Python, with Docker, Kubernetes, Kafka, AWS and Azure. On the AI side I've worked with LangChain, Spring AI, OpenAI, Semantic Kernel, vector databases and MCP. I'm AWS certified in both Solutions Architecture and Machine Learning, and did my MSc in AI with a focus on applied ML.
Experience
AI Enablement Lead / Technical AI Lead
Version 1 — Birmingham, UK (Remote)
09/2023 – Present
DSIT (Dept. for Science, Innovation and Technology), AI Enablement and Adoption · Jan 2026 – Present
- Rotating through government engineering teams for two to four weeks at a time, assessing where they are with AI and getting hands-on with RAG pipelines, prompt engineering and coding assistant setup (Claude Code, GitHub Copilot). Also advising on the bigger picture, from fine-tuning approaches to legacy modernisation.
- Writing playbooks, guides, and worked examples for each team, tailored to their stack and security constraints, so they can carry on without needing us back.
- Wired automated quality gates, model evaluation steps, and compliance checks into teams' existing CI/CD pipelines.
- Feeding trial findings back to senior DSIT stakeholders, which has shaped how the department is approaching coding assistant procurement.
HMCTS and Enterprise AI Delivery · Sep 2023 – Dec 2025
- Built the HMCTS Query Response AI Assistant (Java, Spring AI, Azure OpenSearch, Azure OpenAI), a RAG-based legal assistant with semantic search and function calling that runs at around 4,000 queries per hour. Also built the Wood plc Enterprise Chatbot (C#, React, Semantic Kernel, Azure OpenAI), an internal tool with role-based access control so staff could query enterprise data through a conversational interface.
- Set up the RAG evaluation framework and prompt versioning approach for the HMCTS AI programme. Other delivery teams now use it as a standard pattern.
- Shipped core HMCTS platform features in a properly regulated environment, including Query Management, Settle and Discontinue, Judgment Online and Certificate of Satisfaction.
- Led technical direction for a team of six: low-level design, CI/CD governance, mentoring, and being the person escalations land with.
Senior Software Engineer
Version 1 — Birmingham, UK (Remote)
05/2022 – 09/2023
- Designed and built General Applications for HMCTS, replacing a paper-based process with a microservices solution in Java 17, Kubernetes and Azure so citizens and solicitors could raise applications online.
- Involved in architecture decisions, performance work, and production rollouts across a regulated government programme.
- Code reviews, sprint planning, low-level design docs, and mentoring the rest of the team.
Software Engineer (Backend and ML)
VE Global — London, UK (Remote)
02/2021 – 05/2022
- Built and deployed an ML recommendation engine on SageMaker, covering training pipelines, feature engineering, real-time inference and event-driven analytics on user behaviour.
- Delivered APIs and microservices across C#, .NET, Java Spring Boot, and Python. Migrated workloads to Kubernetes and got ML models into production.
- Built Kafka, Lambda, and DynamoDB pipelines to track and process user behaviour; automated model retraining in SageMaker.
- Part of the team that built a batch sentiment analysis system using Spark on AWS EMR, including a custom BERT model deployed on SageMaker.
Software Engineer (Full Stack, Java)
Accenture S.A — Bilbao, Spain
08/2016 – 10/2020
- Part of a team on a long-running energy sector engagement, covering everything from requirement analysis and impact assessment through to delivery.
- Backend services in JEE and Spring with DB2 and MyBatis, RESTful APIs, IBM MQ messaging. Some frontend work in HTML5, jQuery, and CSS3.
- Estimation, client sessions, technical docs, CI/CD with Jenkins, SonarQube for code quality. Also mentored a junior team member.
Earlier Roles: Full Stack and Mobile Development
SMT S.L. / Sherpa S.L. / BeClever S.L. / Anboto Europe S.L. — Bilbao, Spain
2012 – 2015
- Full-stack and mobile work across Java, Android, .NET, AWS and Azure, including web apps, IoT-driven mobile apps and backend services in JEE, Spring and Hibernate.
- Worked on Sherpa, one of the earlier Android virtual assistants in the Spanish market. Migrated its backend from VB.NET to Java.
- Managed virtual infrastructure (VMware vCenter, vRanger, NetBackup) for several clients, plus performance monitoring with Foglight and Nagios.
Projects
JIRA AI Agent with Spring AI and MCP
An AI agent that creates and manages JIRA tickets through natural language. Built using Spring AI and Model Context Protocol (MCP), it shows how an LLM can plan actions, call external tools, and carry out multi-step workflows. Includes both the MCP server (JIRA tool provider) and the BA-Copilot client.
Writing
Managing AI Projects in the Real World: Beyond Traditional SDLC
How do you run AI projects when the requirements shift, the outputs are non-deterministic, and nothing behaves like traditional software? Practical thoughts on adapting delivery for LLM-powered systems.
From Greedy to Genius: Understanding Decoding Strategies in Large Language Models
Under the hood of how LLMs decide what to say next, covering greedy decoding, beam search, top-k, top-p, temperature and when any of it actually matters.
Getting the Most Out of AI Coding Agents: Key Lessons from Using Windsurf and Cursor
What I've actually learned from using AI coding agents day to day, including what works, what doesn't and how to build habits that get you consistent value.
Enhance Your AI Agent's Capabilities Using Functions in Spring AI
How to extend Spring AI agents with real-time function calling, including wiring up external APIs, defining function schemas and a worked example building a live weather agent with Llama and Spring AI.
Building a JIRA AI Agent Using Spring AI and MCP
What agentic systems are and how to build one, using Spring AI and MCP to create an AI agent that actually does things in JIRA.
Skills & Certifications
Technical Skills
Backend & Architecture
AI/ML & Data
ML Engineering
AI Enablement
Cloud & DevOps
Education
MSc in Computational Engineering and Smart Systems (AI/ML)
2017–2020University of the Basque Country (UPV/EHU), Bilbao, Spain
Research focused on Parkinson's disease detection using multiple ML model architectures (published). Practical experience with PyTorch, model training, fine-tuning, and evaluation.
Master in Software Development and Integration (JEE, Android)
2013University of Deusto, Bilbao, Spain
Software Engineering in Management
2012University of Deusto, Bilbao, Spain
Get In Touch
Always happy to hear from people, whether it's about a project, a role, or just a conversation about AI.