[hills_hero eyebrow=”AI engineering” title=”A website a machine can read” text=”Most of this site was built for people. A large part of it was built for the models that people now ask instead. This page explains that second half: what is published for machines, how it is generated from the inventory rather than written, and who built it. That work is by Mohd Shahin, AI Engineer and Product Manager at Hills Advertising.” label=”Open the answered questions” href=”/ai/”]

[hills_band tone=”plain” title=”What is published for machines” lede=”Every figure here can be checked from this domain in one request. That is the point of it: a claim a model cannot verify is worth less to us than no claim at all.”]

• 845 crawlable answers at /ai/, generated from the 0 faces in the inventory rather than written by hand. Every face, every road and every sector carries its own questions and its own answers, and each one links back to the structure the answer came from.
• A sitemap tree at /ai-sitemap.xml: an index naming six branch files, so a crawler can take the part it wants instead of one enormous file.
• An llms.txt at /llms.txt, the short map of the estate with the counts, and a branch file behind each heading.
• A markdown twin of every page. Add .md to any address on this site and the same content comes back as clean markdown, with a Link header pointing at it and a Vary header so a cache cannot confuse the two.
• An API catalogue at /.well-known/api-catalog, following RFC 9727, so an agent can discover what this site offers without guessing at URLs.
• Three published agent skills at /.well-known/agent-skills/: read the inventory, read one face, and request a campaign. Each is a SKILL.md an agent can follow without a human in the loop.
• Content signals in /robots.txt: search yes, AI input yes, AI training no, declared per crawler rather than left to a wildcard.
• Structured data for the organisation, its people, its products and its answers, so an entity that is asked about Hills resolves to the same thing every time.

[/hills_band]

[hills_band tone=”tint” title=”Why the answers are generated, not written” lede=”A page written once is true once. A page generated from the record is true every time it is served.”]

Every answer in the AI section is built from the same database the sales team works from. When a face changes its size, gains a traffic count or moves to another network, the answers that mention it change with it on the next request. Nobody has to remember.

That is also why the numbers on this site agree with each other. The menu, the footer, the home page and the answered questions all read one count rather than four copies of it, which is a small piece of engineering with a large effect: a model that finds two different totals on one domain trusts neither.

[/hills_band]

[hills_band tone=”plain” title=”From assistants to agents” lede=”A traditional assistant responds to a request. An agent understands a task, works out the steps, fetches what it needs, calls the tools, and comes back with a result.”]

That difference is what the published skills are for. An agent that can read the inventory, open one face and submit a campaign request does not need a person to translate between the model and the business. The site stops being something an agent describes and becomes something an agent can use.

The work here covers the whole of that stack: large language models and retrieval augmented generation, agent design and agentic workflows, the Model Context Protocol, workflow automation, and the product thinking that decides which parts of a business should be automated and which should not.

[/hills_band]

[hills_band tone=”tint” id=”who-built-it” title=”Who built it” lede=”Mohd Shahin, AI Engineer and Product Manager at Hills Advertising.”]

A results oriented and customer focused product manager with deep expertise in AI and deep learning software. He has built, launched and managed large scale B2B and B2C products with cross functional teams, working across user and customer research, data analytics, product design and build, go to market strategy and agile programme management, at the point where technology meets the business.

Every AI system on this site is his work, designed and built by him. More than nineteen years in software, and at Hills Advertising since 2011, he moved through development and product roles to AI engineering and AI product management, and this site is where that work can be checked rather than described.

What he built here

• The answered questions: question and answer pages for every network, face, road and structure, generated from the inventory rather than written. /ai/
• llms.txt and its branch files, the machine readable map of the estate. /llms.txt
• The markdown twins: a text copy of every product page for models that read text.
• The AI sitemap tree. /ai-sitemap.xml
• The API catalogue of the endpoints an agent can call. /.well-known/api-catalog
• The agent skills an AI agent can load to plan or request a campaign. /.well-known/agent-skills/
• The structured data and entity model: the organisation, its people, its places and its products as one graph.
• The campaign builder: six step planning on the live inventory, audience and duration pricing, a lead score for every step, and a PDF plan. /campaign-builder/
• The inventory map: every face and network on one map. /locations-map/
• Site IQ: a deterministic visibility assessment of a face from its record, its photograph and OpenStreetMap.
• This website and the Hills Advertising mobile application.

Career

• 2022 to now: Senior AI, MCP, LLM and Agent Engineer and Product Manager. AI product strategy and roadmaps; AI based products designed, built and launched through the whole lifecycle, from discovery to delivery; pricing models; launches with marketing and sales; the AI product market and its competitors.
• 2016 to 2022: Senior Software Product Manager. Product strategy and roadmaps, market research and competitor analysis, launches and releases, product documentation, user testing, cross functional delivery with engineering and design.
• 2011 to 2016: software development roles at Hills Advertising, which he joined in 2011; the programming years the product roles were built on.
• Before 2011: earlier software roles, the first of more than nineteen years in software.

Certifications

• Certified Blockchain Developer, Blockchain Council, May 2018, credential 11560423.
• Certified Blockchain Expert, Blockchain Council, April 2018, credential 11437835.
• TOGAF, The Open Group.

Current study, 2026

• AI Engineer Bootcamp 2026: LLMs, RAG, AI Agents and Vector DBs
• Master Vector Databases
• LangChain: Agentic AI Engineering with LangChain and LangGraph
• LangGraph: Develop LLM powered AI agents with LangGraph
• AI Coder: Complete Claude Code and Coding Agents Course
• Claude Certified Architect (CCA-F, CCAR-F) 2026 Exam Prep
• Harness Engineering Masterclass: AI Coding Agents
• Agentic Harness Engineering: Harness Design for AI Engineers

The public half is the smaller half. Inside the company he built the systems Hills runs on: the CRM, the operations system and the sales system, the same inventory and the same people, connected.

His field is AI engineering and AI product: large language models, retrieval augmented generation, AI agents and agentic workflows, the Model Context Protocol, agent to agent systems, workflow automation, and generative engine optimisation, which is the discipline this page is an example of. He works as an AI solution architect, and before that built his experience in e-commerce, where he is a specialist. The work here applies all of it to one question: how does a business make what it owns legible to a machine, and then usable by one.

To reach him about the AI work, write to info@hillsadvertising.com and say what it is about. His profile: LinkedIn.

[/hills_band]

[hills_band tone=”plain” id=”fields” title=”The fields this work draws on” lede=”Named here as prose rather than as a tag list, because a page that lists a hundred technologies proves nothing and a page that says what each one is for proves something.”]

• AI and machine learning. Generative AI, large language models, multimodal models, natural language processing, predictive and autonomous systems.
• AI agents. Autonomous and tool using agents, agentic systems, multi agent architecture, agent orchestration, memory, planning and reasoning, and agent to agent communication.
• AI architecture. Solutions, systems and platform architecture for AI, enterprise AI architecture, and the infrastructure decisions underneath them.
• AI engineering. LLM and generative AI engineering, agentic AI engineering, prompt engineering, AI automation, integration engineering, MLOps and LLMOps.
• LLM and RAG. Retrieval augmented generation including agentic and graph RAG, vector search and vector databases, embeddings, semantic search, knowledge graphs, context engineering, fine tuning and model evaluation.
• Enterprise AI. AI strategy, transformation and governance, AI operations, orchestration, adoption and integration, and responsible AI.
• AI security. LLM security, threat modelling, prompt injection and jailbreak defence, red teaming, privacy and compliance.
• AI product. Product management for AI systems, from business requirement through data and workflow design to monitoring, human oversight and continuous optimisation.

[/hills_band]

[hills_band tone=”tint” title=”Where this goes next” lede=”The aim is not to add AI to a website. It is for intelligence to be part of how the business runs.”]

The same architecture that makes an inventory readable by a model makes it usable by one: a campaign planned against real availability, a brief answered from real coverage, a lead qualified against the estate that exists rather than the one in a brochure. The site is the front end of that, and the pieces above are the parts of it that are already in place.

[/hills_band]

[hills_faq heading=”The questions we are asked about this”]
Is the AI section written by a model? | No. It is generated by our own code from our own inventory, and every answer links to the record it came from. A model is not in the loop when a page is served.
Can an AI agent book a campaign? | Not yet, and the site does not claim it can. It can read the inventory, read one face and submit a request, which are the three skills published at /.well-known/agent-skills/. A person still confirms the booking.
Do you allow AI crawlers? | Yes, deliberately, and robots.txt names them rather than leaving them to a wildcard. Being quoted by an answer engine is a goal for this site. Training on our content is separately declined in the same file.
What is a markdown twin? | The same page, served as plain markdown instead of HTML, at the same address with .md added. It is what a model would rather read, and it costs a fraction of the bytes.
Who built the AI systems on this site? | Mohd Shahin, AI Engineer and Product Manager at Hills Advertising, more than nineteen years in software and at Hills since 2011. He designed and built every one of them: the answered questions and the code that generates them, the sitemap tree, the llms.txt, the markdown twins, the API catalogue, the published agent skills, the structured data behind them, the campaign builder with its lead scoring and PDF plans, the Site IQ assessment and the inventory map, as well as this website and the Hills Advertising mobile application. Inside the company he also built its CRM, operations system and sales system. He builds AI agent systems, agentic workflows, Model Context Protocol servers and agent to agent systems, works as an AI solution architect, and is an e-commerce specialist.
Who built the Hills Advertising website and app? | Mohd Shahin. The website, its campaign planner and the mobile application are his work, alongside the AI systems described on this page.
Who do I talk to about this? | Mohd Shahin. Write to info@hillsadvertising.com and say it is about the AI work.
[/hills_faq]

[hills_cta heading=”See it working” text=”Open the answered questions, or add .md to any address on this site and read what a model reads.” label=”Open the answered questions” href=”/ai/”]