Generative AI and data · UK
I build it. Then I make sure it gets used.
Generative AI systems, and the data underneath them. Building it is the half everyone talks about — adoption, governance, the business case, and someone still owning it a year later is where the value actually shows up. I work across both.
25 years delivering inside extremely regulated environments — safety-critical and audited systems, UK.
Most people arrive with one of these. Start wherever you are.
- 01
Getting value out
The licences went out months ago and nobody can say what changed.
More → - 02
What to build
You have been told to "do something with AI" and nobody agrees what.
More → - 03
Data for AI
The model is fine. Your data is not ready for it.
More → - 04
AI engineering
The demo worked. It died on the way to production.
More → - 05
AI governance
People are creating agents faster than anyone can approve them.
More → - 06
AI teaching
Your teams were handed the tools and never taught the thinking.
More → - Not sure?
Then it is probably the first one.
Send me a paragraph about what is on your desk. I will tell you honestly whether I can help.
Start a conversation →
◆ Where it usually stalls
The pilot works. It never becomes how the work gets done.
Six months on, the demo is still a demo. What stopped it is almost never the model.
A pilot that answers well on a clean question, in front of the people who commissioned it.
Data nobody curated, permissions nobody audited, a risk owner who never signed, and no one who can say what it saved.
I join the organisation, not the supplier list.
Most engagements start with me becoming part of the team for a period, rather than reporting on it from outside. I work on the organisation’s behalf, inside its constraints, its approval routes and its culture.
That asks for adapting quickly — learning how decisions really get made, what can and cannot be said in which room, and who will still be living with the thing long after I have gone. It is the reason the second half of my job is possible at all. You cannot get something adopted from the outside of a building.
The industry has started calling this a forward-deployed engineer. It is one of the ways I work, not the only one.
◆ Two pillars
Where to look next.
Everything here is open and dated. Follow whichever one fits what you are trying to work out.
Research
Independent write-ups on secure GenAI, governance and data readiness — grounded in primary sources, no hype, uncertainty shown.
Read the research →if you want the receiptsProjects
Real, running systems and honest post-mortems. Methodology over outcomes — including the experiments that found no edge.
See the work →Free to read, no sign-up.
Everything I have written up while learning this stack myself. Published because it is more useful out here than on my own disk.
AI-103
Developing AI Apps and Agents on Azure
38 revision cards, each one on its own page with a plain-English write-up underneath. Mapped against the official skills outline.
Start at card 01 →- AB-100
Agentic AI Business Solutions Architect
Forty-four revision cards, each one explained in plain English.
Open → - AI-500
Multi-Agent AI Solutions Expert
Forty revision cards, each one explained in plain English.
Open → - Course
Python for AI Engineers
Fifteen chapters, from the language basics to a full working app.
Open →
◆ Selected work
Things I built and can defend.
Each one is a real system with a real outcome — and the reasoning is public.
AI Governance Journey
A plain-English AI governance framework in twelve stages, with automated pipelines that watch the major regulations and raise a pull request when one moves.
Defence-AI Radar
A self-updating timeline of defence AI across the UK, Europe and the US, built from public sources only — with no language model anywhere in the pipeline, so every entry traces to a source.
Self-improving agent platform
The agentic system I actually run every day. It mines its own sessions for durable lessons and writes them back into memory, and earns new capabilities through single-purpose allowlisted scripts rather than wider permissions.
Python for AI Engineers
A hands-on course for engineers moving into AI work. Real code, no unexplained acronyms, free to read on this site.
◆ Flagship research hub
The signature piece
The AI Engineering Pipeline
The whole path, mapped end to end: idea → design → build → deploy → monitor. The frameworks, the SDKs, and the trichotomy I keep coming back to — Prompt vs Context vs Loop engineering — plus a working shootout of Claude Code vs Codex vs OpenCode.
Work with me
Move your GenAI from pilot to production — and make it land.
Twenty-five years of getting systems into production in places where being wrong matters. If you are stuck between “the tech works” and “it is still not in production”, that gap is the work.
- Email meanandbg@gmail.comA paragraph about what is on your desk is plenty.
- Message me on LinkedInlinkedin.com/in/anandbgFine if you would rather see who I am first.
- Book a callHalf an hour, freeWhen you already know you want to talk it through.
Best fit: heavily regulated organisations, one hard adoption problem, a real decision on the table.