Instagram story viewer> @chrisgarlick.ai> Posts

Chris Garlick | AI Architect

@chrisgarlick.ai

4.4K

followers

142

following

🧠 AI Engineering & Platform Architecture.
⚙️ Automating business data & workflows.
💰 Optimise LLM pipelines to save on bills.
👇 Book an audit now

POSTS STORIES REELS TAGGED

GPT-5 just helped solve a science problem that had a lab stuck for three years.

OpenAI published the account: working with immunologist Derya Unutmaz, GPT-5 Pro simulated an experiment on how glucose shapes T cells and correctly predicted a result he hadn’t published yet, so it couldn’t have read the answer online. It also helped a mathematician close a 40-year-old open problem. The catch: humans still ran and verified the real lab work. The AI sped up the thinking, it didn’t replace the scientist.

The shift: AI is starting to find answers, not just fetch them.

Source: openai.com
(25 Jun 2026)

#AInews #OpenAI #GPT5 #AIresearch #LLM by @chrisgarlick.ai

4

0

4 months ago

Onboarding one client costs your practice 110 minutes and a week of chasing emails. Every client. There’s a better way.

Most of what happens between signing a new client and starting their work is admin a machine should be doing. Collecting documents. Chasing for the third time. Running ID and AML checks. Screening sanctions and PEP lists. Drafting the engagement letter. None of it needs a partner’s judgement, yet it usually lands on a senior person’s desk.

So here’s the honest breakdown. Where the 110 minutes actually goes. The four chunks AI genuinely automates. And the one step you should never hand over (the risk decision stays with your MLRO, every time, under MLR 2017).

Swipe through, then save it for your next practice meeting.

The numbers are real and sourced: 110 minutes across a week of waiting drops to roughly 15 minutes in one same-day session. AML checks run from around £2.10 per check. And the audit trail you get as a by-product is exactly what your supervisor wants to see.

Automation does the 90 minutes of admin so your team spends its time on the 20 minutes of actual decision. That’s the whole game.

Save this for your next practice meeting. Want to know where it would pay off in your firm? Run a free audit at chrisgarlick.com/audit. And tag a practice manager or partner who is still chasing onboarding documents by hand.

.
.
.

#onboarding #ai #aitools #business #businessowner by @chrisgarlick.ai

1

0

4 months ago

5

0

4 months ago

5

0

4 months ago

UK agencies are losing 200 hours a month to client reports clients barely read in full.

That’s the brutal version. The polite version is “we just spend too long on assembly and not enough on analysis.” Same problem.

Here’s the bit nobody pitching automation tools tells you. The off-the-shelf AI summaries (the ones built into AgencyAnalytics, Databox, the rest) save you about 10 minutes per report. Useful, but they still sound like every other agency using the same tool. They are not a substitute for an agency analyst.

The agencies actually pulling ahead in 2026 are building a thin custom layer on top of the off-the-shelf stack. Same connectors. Same dashboards. Custom AI commentary in the agency’s voice, against a structured fact pack, with a fact-check pass that catches hallucinations before they go to clients.

That’s what flips a 90-minute write-up into a 4-minute review. That’s what makes the ROI a real story rather than a productivity nice-to-have.

The 10-slide guide walks through:
1. Why the “human touch” argument has aged badly
2. The 4-layer reporting stack (connectors, warehouse, commentary, delivery)
3. The two-pass pattern that gets AI commentary past the slop threshold
4. When off-the-shelf is genuinely enough
5. The build-vs-buy decision matrix for each layer
6. What an 8-account-manager agency actually recovers (170 hours, ~£17k/month)

Save this for your next ops review. Read the full guide at chrisgarlick.com/article/ai-reporting-automation-agencies. Tag an agency owner who needs to see this.

#aiintegration #marketingagency #agency #client #reporting by @chrisgarlick.ai

0

0

5 months ago

“Just use Claude” is the most common UK AI advice of 2026.

It’s also wrong often enough that giving it as the default is bad advice.

The shape of the market changed in 2025. Open-source closed the gap. Self-hosting flipped cheaper at moderate scale. Compliance teams woke up. “Just use Claude” now costs UK firms five figures a year when better answers exist.

Five reasons your default LLM is wrong:

1. Your client data is leaving the country
2. Costs flip at moderate scale ($47k to $8k for one UK fintech)
3. Anthropic killed Claude 3. Yours is next.
4. You can’t really fine-tune
5. Open-source already caught up. Qwen 3.5 just passed 700M Hugging Face downloads.

But sometimes Claude IS the right answer. Under 5M tokens a month, varied workloads, no engineering capacity, no regulated client data. Hit all four? Use Claude. Move on.

For everyone else, the answer is more nuanced. Swipe through for the 4-question picker that walks you to your model.

Save this for the next time someone says “just use Claude”.
Get the full LLM Cheat Sheet at chrisgarlick.com/resources.
Tag a partner, FD or IT lead who needs to see this.

#ai #claude #aimodel #aidevelopment #webdev by @chrisgarlick.ai

4

0

5 months ago

Ask ChatGPT about your own contracts and it’ll confidently invent the answer. That’s the problem RAG solves.

RAG stands for Retrieval-Augmented Generation. Three steps. Retrieve the most relevant chunks of your own documents. Augment the user’s question with those chunks as context. Generate the answer using only that context.

The model summarises. It doesn’t recall. That single architectural change is what cuts hallucinations.

Across 847 production deployments, RAG cuts hallucination rates by a median 71% on domain-specific queries. For summarisation tasks specifically, grounded retrieval drops hallucinations below 2%.

How it actually works, in 4 steps. No jargon.

One. Ingest. Break every document your AI should know about into smaller chunks. PDFs, contracts, Notion pages, emails. A paragraph at a time.

Two. Embed. Each chunk becomes a vector. A list of numbers representing what the chunk “means” so you can search by semantic similarity, not exact keywords.

Three. Store. The vectors live in a vector database. pgvector if you already use Postgres. Qdrant or Pinecone for scale or strict latency.

Four. Retrieve plus generate. Query embeds, find the closest chunks, attach them to the question as context, send the package to an LLM. The LLM answers from your documents.

What a RAG build typically costs across the UK market in 2026. These are industry benchmarks, not my rate card.

Single-purpose RAG. £15k to £40k upfront. Two to six weeks. £100 to £500 a month to run.
Multi-corpus with admin UI. £40k to £80k. Four to eight weeks. £300 to £1.5k a month.
Multi-agent with workflow integration. £80k to £150k. Eight to sixteen weeks. £500 to £3k a month.

Source: UK AI development cost benchmarks 2026 (openkit.co.uk).

Most UK SMEs starting today should be in tier one. Tier two earns its keep when accuracy and auditability matter. Tier three is for serious operational integration where AI is embedded in the operational software itself.

If you’d like to find out more about RAG, hit the link in my bio and head over to articles to read more.

#ai #aitools #aimodel #aicommunity #artificalintelligence by @chrisgarlick.ai

5

0

5 months ago

A front end project that increments a number when the button is clicked.

This can be very useful for things like web based games, when you may need to increment the item now owned by the user.

Tell me below how you’ve been using JavaScript in front end projects 👇🏼👇🏼

@coder_brain @c4code @coding_.master @coding_gyan_ @coding.batch @codergallery @webdevelopment24x7 @webdevelopment.projects @webcodingdev @javascript.js

#coding #codinglife #codingbootcamp #coder #coderlife #codegoals #coderpower #javascript #javascriptdeveloper #javascripttutorial #js #html #htmlcoding #css #csstricks #frontenddeveloper #frontendfriday #frontend #fullstackdeveloper #fullstackdev #gamedev #gamedevelopment #webdeveloper #webdevelopment #webdev taken in London, United Kingdom by @chrisgarlick.ai

72

7

5 years ago

NEW BLOG - OUT NOW - LINK IN BIO! In this blog we look at the Tuple data type and how it differs from other data types.

If you’re unsure on what they are, when to use them or why we use tuples, then be sure to read the blog for more!

How often do you use Tuples?

@python.learning @pythoncommunity_ @python___developers @python.science @codergallery @coder_brain @coding_.master @codinganddecoding

#coding #codechallenge #codingblog #codingbootcamp #learntocode #pythonprogramming #pythoncode #pythontips #data #datascience #dataanalytics #webdevelopment #computerscience #computersciencestudent #softwareengineer #softwaredevelopment #gamedev #developer #developerlife #devops #git #github #aws #strings #bool #lists #dictionary #set #number taken in Newcastle upon Tyne by @chrisgarlick.ai

60

3

5 years ago

I have been spending some of my time learning more about JavaScript and how we can use them to make our websites more interactive and requiring less loading time.

In this webpage we use JavaScript to hide the information we don’t need to show until a user clicks on a button, this allows for our content to load but only be visible depending on what page they are viewing.

Tell me below if you have any experience with JavaScript 👇🏼👇🏼

@webcodingdev @webdevelopment24x7 @webdev_guru @webdevelopment.projects @htmlhints @frontendjoe @frontendwebdevelopers @coder_brain

#coding #codingforweb #codinglife #codingbootcamp #coder #coderlife #learntocode #programmerlife #programmer #softwaredeveloper #softwareengineer #websitedesign #website #frontend #frontenddeveloper #frontendchallenge #html #html5 #css #cssinjs #js #javascript #javascriptdeveloper #javascripttutorial #javascript30 #beginnerscode taken in London, United Kingdom by @chrisgarlick.ai

95

8

5 years ago

CSS plays a massive part in not only making your website design more sleek and professional, but it can also improve the users experience when using your site.

In part 1 we look at what CSS is, how we can use it to style html and where we can place our code!

Tell me below how you have been finding CSS 👇🏼👇🏼👇🏼

@frontendjoe @frontendwebdevelopers @frontendsimply @the_frontend_developer @frontendeverything @webcodingdev @webdev_guru @coding_.master @codergallery @coder_brain

#coding #codinglife #codingbootcamp #learntocode #webdevelopment #webdeveloper #webdev #webdeveloperslife #frontend #frontendfriday #frontenddeveloper #frontendchallenge #frontenddev #css #css3 #csstricks #cssanimation #htmlcss #ui #ux #webdesign #buildtheweb #computerscience #softwareengineer #softwaredeveloper #fullstackdeveloper #fullstacklearning #fullstackdev by @chrisgarlick.ai

76

9

5 years ago

When replicating a real world object for our databases can be very tricky and confusing for beginners. Luckily using Django this can be a lot easier!

Not only are we able to use the terminal AND the admin user interface to add/manage our database, but we can easily create a variable name to use in our HTML to access our models!

Tell me below what Django projects you’ve been working on 👇🏼👇🏼👇🏼

@codecrookshanks @webcodingdev @webdev_guru @webdevelopment24x7 @coder_brain @coder__bhai @codergallery @coding_.master @sparshcodes @frontendjoe @frontendeverything

#coding #codinglife #codingbootcamp #codingproblems #learntocode #python #pythondjango #pythonprogramming #pythonexplainedtome #webdevelopment #webdeveloper #website #websitelaunch #singlepagewebsite #frontenddeveloper #frontend #frontendchallenge #backenddeveloper #backend #backenddevelopment #fullstack #fullstackdeveloper #fullstackdev #buildtheweb #https #ssl #cookies #sessions #models taken in London, United Kingdom by @chrisgarlick.ai

72

7

5 years ago