Personalized for you

Oh hey, Sitemapxml! So glad you're here!

I've put together some of my favourite projects just for you.

See the work

The work

Built systems, not just decks.

Below: what I'm most proud of shipping.

You're looking at it

Claude Code

A site that knows who's walking in.

Dynamic. Instant. Intentional.

The URL isn't decoration — it's the conversation. Change the slug, change the message. Sanitized input, heuristic classification, zero APIs, and a personalized metadata title for whoever lands on it.

Dynamic routing. Next.js App Router with a single [slug] segment.
Safe by default. Every slug is sanitized before it touches the DOM.
Heuristic logic. People, companies, and generics — one pure function.

yourname.com/stripe

Personalized

A quick note for the team at Stripe

Here's a curated look at how I think about AI, marketing, and transformation.

Case study

A site that knows who's walking in.

Dynamic. Instant. Intentional.

The problem

A static resume or portfolio speaks the same way to every viewer. When you want to signal intentionality to a specific recruiter or team, there's no cheap way to do it — and a generic homepage is the opposite of intentional.

The solution

A Next.js site where the URL slug personalizes the hero, intro, and sign-off. Send one link — yourname.com/stripe — and the recipient lands on a page that reads like it was written for them, because every instance of it was.

How it flows

Input

URL slug
/stripe
/john-doe

Sanitize

Strip unsafe chars
Title-case
Cap length

Classify

Known company?
Hyphen?
Short word?

Render

Hero
Intro
Sign-off

Steps to solve

01
Define the slug contract. One URL segment. /{name} is the only variable that matters — everything downstream derives from it.
02
Sanitize and format. Pure functions strip unsafe characters, cap length, and title-case the result. No regex in the UI layer.
03
Classify heuristically. Known-companies list plus simple rules (hyphen → person, short single word → person, rest → generic). No APIs, no ML.
04
Render three variants. Person, company, and generic copy paths. All other UI pulls from a single content object returned by one pure function.
05
Extend everywhere. <title> tags, intro sentence, CTA sign-off — the same name flows through the whole page, not just the hero.

Inputs & context

URL slug. The only input that matters.
Known-companies list. Small Set<string>, easily extended.
Classification rules. A handful of lines, no dependencies.

Outcome

One link, infinite recipients, zero per-recipient effort. A three-minute deploy turns every outbound into a tailored landing.

Visuals

Screenshot placeholder

The /stripe variant — company treatment with team-level copy.

Screenshot placeholder

The /sally variant — warm, name-first person treatment.

CoGPT · Custom AI Systems

ChatGPTCustom GPT

A custom AI system built around your context.

Your business, brand, and strategy preloaded for faster, more consistent work.

Context-loaded. Business strategy, voice, audience, and offers preloaded — no re-explaining.
Workflow-native. Built around recurring tasks, approval steps, and real output formats.
Prompt-independent. Users get strong results without being expert prompters.
Continuously refined. Tested against actual tasks and tightened until it fits naturally into everyday work.

CoGPT

Context loaded

Strategy

PositioningOffersVoice

Context

AudienceWorkflowsSource docs

Output

DraftsIdeationExecution

Draft the landing page for…

Case study

A custom AI system built around your context.

Your business, brand, and strategy preloaded for faster, more consistent work.

The problem

Most AI tools start empty. They may be powerful, but they know nothing about the business using them. Teams are left to re-explain context, rewrite generic drafts, and prompt their way toward something usable. In most businesses, the real value lives across documents, notes, messaging, offers, workflows, and people's heads. Without that context structured into the system, AI stays generic, inconsistent, and underused.

The solution

CoGPT is a custom AI system built around the real context generic tools lack: your business model, positioning, audience, offers, voice, priorities, source material, and workflow logic. Each build is shaped around a real job to be done, with clear inputs, structured instructions, and practical output patterns. The result is not just a chatbot. It is a usable system that helps teams think, draft, and execute faster without starting from zero every time.

How it flows

Inputs

Business strategy
Offers and services
Audience insights
Voice and messaging rules

Load

Capture business context
Encode positioning logic
Organize source material
Define output standards

Guide

Apply voice rules
Structure tasks and prompts
Add workflow guardrails
Reduce generic output

Output

Better first drafts
Faster content execution
Stronger consistency
Less rework and repetition

Steps to solve

01
Capture the business foundation. Clarify the offer, audience, positioning, differentiators, and voice the assistant needs to understand.
02
Pull out repeated context. Extract the information normally repeated across meetings, revisions, onboarding, and day-to-day execution.
03
Identify high-value use cases. Map where the CoGPT should create leverage — whether drafting, ideation, refinement, strategy support, or recurring internal tasks.
04
Build the system logic. Structure instructions, workflow steps, messaging rules, and output expectations around how the business actually works.
05
Embed strategy into execution. Make the assistant useful not only for writing, but for shaping direction and producing stronger outputs from the start.
06
Reduce reliance on prompt skill. Build enough intelligence into the system that users can get strong results without being expert prompters.
07
Refine through real use. Test against actual tasks, tighten outputs, and improve reliability until the build fits naturally into everyday work.

Inputs & context

Business strategy. Core offer, priorities, differentiation, and how the business creates value.
Brand voice. Tone, messaging patterns, language preferences, and positioning cues.
Audience context. Who the business serves, what those people care about, and how the business should speak to them.
Workflow requirements. The recurring tasks, approval steps, and output formats the assistant needs to support.
Source material. Existing content, notes, documents, frameworks, and examples that ground the system in real business context.

Outcome

Stronger first drafts, faster execution, and less reinvention — because the assistant starts with the business context already built in.

Visuals

Screenshot placeholder

Business strategy, voice, and workflow logic loaded into a CoGPT instead of recreated from scratch each time.

Screenshot placeholder

A stronger starting point for drafting, ideation, and execution — with context and structure already built in.

Little Rebels Music & Development

Claude Cowork

An Instagram engine built for a real small business.

Strategy Bible in. Monthly calendar out.

A content system for an early-childhood music studio. It plans monthly calendars, drafts captions in the founder's voice, composes branded image and video posts, and publishes to Planable — all from a few commands.

Strategy Bible. One source of truth for voice, tone, and brand.
Content Cornerstones. Balanced mix across awareness, leads, sales, and retention.
Reel Composer. 1080×1920 Reels composed from raw footage with motion text.
Publish Pipeline. Google Doc → Sheet → Zapier → Planable drafts.

Tiny humans.

Big feelings.

Real music.

♡ 2.4k💬 87↗ 312

Case study

An Instagram engine built for a real small business.

Strategy Bible in. Monthly calendar out.

The problem

A real small business was spending 6+ hours per week planning, drafting, and composing Instagram content. Voice drifted across posts, the owner was burning out on social media, and none of it was being measured against the business's actual funnel.

The solution

A plugin ecosystem that captures the brand's Strategy Bible, plans monthly calendars using a four-cornerstone framework, drafts captions in the owner's voice, composes branded image and video posts, and drops everything into Planable as drafts — ready to review.

How it flows

Strategy

Strategy Bible
Audience
Cornerstones

Plan

Monthly calendar
Weekly batches
Cornerstone mix

Create

Caption drafts
Image posts
Video Reels

Publish

Google Sheet
Zapier
Planable drafts

Steps to solve

01
Capture the Strategy Bible. Voice, tone, audience, offers, and brand rules live in one structured document the plugins all read from.
02
Plan with Content Cornerstones. Balance awareness / leads / sales / retention per month — no more accidental 30 promo posts in a row.
03
Draft captions in-voice. Every caption is generated against the Strategy Bible so the brand sounds like the brand.
04
Compose the media. Branded image posts via Canva templates, Reels composed from raw footage with motion text and gradients.
05
Publish through one pipeline. Google Doc → Google Sheet → Zapier → Planable. The owner reviews drafts in a single surface.

Inputs & context

Strategy Bible. Voice, tone, audience, offers, differentiators.
Content Cornerstones targets. Monthly mix across four objectives.
Raw media assets. Photos and video clips from classes and events.
Planable workspace. Destination for draft posts, scheduling, and approvals.

Outcome

6 hours/week → ~45 minutes of owner review. Voice consistency maintained across 100% of posts.

Visuals

Screenshot placeholder

Monthly content calendar rendered against the Cornerstone framework.

Screenshot placeholder

Reel composer output — 1080×1920 with motion text and brand gradient.

Sawmill Structures · Meta Ads Dashboard

Custom Web AppClaude CodeMeta API

A client portal that always knows which ads are running.

One link → every active creative, copy variation, and approval — updated every morning.

An always-on dashboard built for a Meta advertising client. It pulls live campaign data, matches each ad to its source creative, and surfaces copy variations, performance metrics, and a creative review queue — all behind a secure client login and refreshed automatically every night without anyone touching a thing.

Live creative grid. Every active ad across all funnel stages, displayed with its image or video creative.
Copy variations panel. All headlines, body copy, and descriptions in one place — deduplicated and organized by funnel stage.
For Approval tab. Drop a file in the right place. It appears on the dashboard overnight. No emails, no manual uploads, no developer required.
Archive tab. Retired creatives preserved for reference — visible but clearly separated from what's live.
Secure client login. Password-protected access so the client always has a private, up-to-date view of their account without needing Ads Manager access.

Sawmill Structures

Live

Top Funnel

Live

Mid Funnel

Live

Bottom Funnel

Review

Retarget

ActiveApprovalArchive

Case study

A client portal that always knows which ads are running.

One link → every active creative, copy variation, and approval — updated every morning.

The problem

The agency managing Sawmill Structures' Meta campaigns had no single place to show a client what was running. Creative review happened over email. Checking which copy was live meant navigating Ads Manager. When new creatives needed approval, it was a back-and-forth of file shares and screenshots. Nothing was archived systematically. Every week the same questions came up: what's live right now? what's waiting for approval? what did we test before?

The solution

A purpose-built client portal — accessible by secure link, refreshed automatically every morning — that answers all of those questions without anyone doing manual work. It connects directly to the Meta advertising platform and the agency's asset storage, assembles everything into a clean dashboard, and updates itself on a schedule. The client sees their creatives exactly as they're running. The agency spends zero time preparing reports or hunting down files. When something new needs approval, it appears automatically. When something is retired, it moves to the archive on its own.

The Manual Process (Before) · 2–3 hours/week

Check What's Running

Navigate Ads Manager campaign by campaign
Screenshot or note active creatives
No single view of all stages at once

Gather Copy

Open each ad individually to read the copy
No deduplication — same text seen multiple times
Easy to miss variations across ad sets

Share for Approval

Export or screenshot new creative files
Send to client via email or Slack
Track responses manually, no paper trail

Reference Past Work

No archive system — retired creatives just got lost
No record of what was tested or why it stopped running

How It Flows (After) · ~5 minutes/week

Upload

Add new creative file to the shared asset folder
No other action required

Review

Open the dashboard link
All active creatives visible by funnel stage
Copy variations surfaced automatically

Approve

New creatives appear in the For Approval tab overnight
Client reviews at their private link — no email thread needed

Steps to solve

01
Built the data pipeline. Connected directly to the Meta advertising API to pull live campaign and ad data — creative assets, copy, and 30-day performance metrics — on a nightly schedule.
02
Solved the creative matching problem. Each ad needed to display its actual source file, not just a low-res platform thumbnail. Built a system to locate and serve the original creative from the agency's asset storage, with automatic matching logic that handles new files without manual configuration.
03
Untangled Meta's copy structure. Meta stores ad copy differently depending on the campaign type. Built parsing logic that handles both standard and dynamic campaign formats, pulling all copy variations and surfacing them in a clean, deduplicated panel.
04
Automated the approval workflow. The For Approval tab populates itself — files added to a designated folder appear on the dashboard overnight, with format detection handled automatically.
05
Built the archive system. Retired creatives are preserved and surfaced in a separate tab, giving the team a searchable history of what has run without cluttering the active view.
06
Secured client access. Implemented authentication so the dashboard is private to the client — accessible by link, no accounts or passwords to manage on their end.
07
Made it self-maintaining. The entire pipeline runs on a schedule with no manual trigger. If data changes in the ad account, the dashboard reflects it the next morning.

Inputs & context

Meta Ads platform. Live campaign, ad, and creative data including copy variants and 30-day performance metrics — impressions, reach, clicks, spend, and video performance.
Cloud asset storage. All source creative files stored and served from the agency's asset library, matched to their corresponding ads automatically.
Automated deployment pipeline. Nightly data refresh commits updated content and triggers a live deployment — no server to manage, no manual publishing step.
Client authentication layer. Secure, low-friction access for the client — verified by email on first visit, no account setup required.

Outcome

Manual creative reporting eliminated. The client has a private, always-current view of every running ad, every copy variation, and every creative pending approval — without anyone preparing it.

Visuals

Live dashboard — every active creative, organized by funnel stage with performance metrics.

Screenshot placeholder

Creative grid with approval status indicators and copy variation panel.

Urban Jars · Meal Prep Automation

Custom Web AppClaude Code

A grocery list that builds itself from weekly orders.

3–4 hours → 30 minutes. Zero missed items.

An order-to-grocery automation tool for Urban Jars Cambridge. It ingests the weekly Wix order export, explodes recipes into ingredients, adjusts for inventory, then emits a menu tally and printable grocery list — grouped by store aisle, scaled to order count.

CSV import. Parses Wix orders — variants, mods, combos — automatically.
Recipe explosion. 566 recipe rows × order qty = exact ingredient totals.
Three-tier grocery. Package counts, raw weights, or flagged for review.
Menu tally. Matches the owner's printed template exactly — same columns, same shorthand.

Urban Jars — Grocery List

92 orders

Vegetables

✓Red Onion · 1 × 4540g

Carrots · 1 × 907g

Kale · 425g

Peppers · 2 × 1130g

Dairy

Feta · 2 × 200g

Gouda · 1 × 400g

Cottage Cheese · 60mL

Grains

Brown Rice · 3 × 900g

Quinoa · 2 × 1800g

Basmati · 1 × 3630g

Case study

A grocery list that builds itself from weekly orders.

3–4 hours → 30 minutes. Zero missed items.

The problem

Urban Jars is a meal prep company offering 40+ weekly menu items — lunch jars, skillets, breakfast parfaits, protein bites, and more. Every week, the owner exports Wix orders and manually translates them into a menu tally and grocery list: counting items, looking up recipes, multiplying quantities, checking inventory, and calculating store packages. With 90+ order lines, customer modifications (GF, dairy-free, no onion), combo packages to unpack, and variant routing (yogurt type, skillet grain base, jar size) — this took 3–4 hours of error-prone manual work every single week.

The solution

A full-stack Next.js web application with a deterministic 10-step processing pipeline. The owner uploads her Wix CSV, and the system parses every order, explodes combo packages into individual items, classifies dietary modifications, maps products to recipes, computes exact ingredient quantities, deducts inventory, calculates store package counts, and builds a formatted menu tally — all matching her existing printed template. No guessing, no estimates. Missing data gets flagged, not silently skipped.

The Manual Process (Before) · 3–4 hours/week

Read & Sort

Open CSV in Excel
Read each row manually
Note variants & mods

Count & Route

Tally each product by hand
Route to correct column
Unpack combo orders mentally

Recipe Lookup

Look up each recipe in spreadsheet
Multiply × order count
Sum across all products

Grocery Math

Check pantry inventory
Calculate packages to buy
Hope nothing was missed

How It Flows (After) · < 30 minutes/week

Upload

Export CSV from Wix
Upload to the app
Pipeline runs automatically

Menu tally generated
Grocery list by store aisle
Print both for the kitchen

Review

Quick scan of the output
Check exceptions flagged
Adjust quantities if needed

Shop

Grocery list in hand
Grouped by store section
Package counts calculated

Steps to solve

01
Parse the Wix CSV. Filters non-food items (eGift Cards), non-paid orders, and extracts the date range for the weekly run.
02
Explode combo packages. Starter Kits and Bundles are split into individual items — a single combo row becomes 5–6 separate order lines.
03
Parse variants & classify modifiers. Size, yogurt, flavour, skillet base extracted from pipe-delimited strings. Customer mods classified into dietary buckets (GF, DF, NO_ONION).
04
Map to recipes & resolve sizes. Wix item names matched to internal product IDs. Recipe size determined by category defaults and explicit variant data.
05
Explode ingredients. Each product's recipe looked up, quantities multiplied by order count. Grain substitution for skillets handled dynamically.
06
Compute grocery list. Ingredients aggregated, inventory deducted, three-tier system applied: package counts → raw weight → flagged for review.
07
Build menu tally & export. Structured tally with column routing, cell consolidation, and modifier annotations. Excel exports and print-ready views generated.

Inputs & context

Wix orders CSV. Weekly export with order number, item, variant (pipe-delimited), quantity, payment status, customer notes, and modification text.
Recipe database (566 rows). Full ingredient breakdown for each product at each size — every ingredient with exact quantity in grams or millilitres per single order unit.
Product catalog (43 products). Lunch jars, skillets, breakfast jars, protein bites, cheesecakes, soups — with size options and Wix name mappings.
Package rules (141 rules). Store package sizes per ingredient (e.g., brown rice = 900g bags, feta = 200g blocks) for calculating "buy X packages" recommendations.
Modifier rules (7 rules). Keyword-to-dietary-bucket mappings: "gluten free" → GF, "celiac" → GF, "dairy free" → DF. Priority-ranked for correct classification.
Ingredient master list (168 items). Canonical ingredients with store section assignments (Dairy, Vegetables, Grains, etc.) and preferred units.

Outcome

3–4 hour weekly task reduced to under 30 minutes. Zero manual calculations. Every anomaly flagged — not silently skipped.

Visuals

Order upload — drag-and-drop the weekly Wix export.

Grocery list output — grouped by aisle, with review flags for anomalies.

Power Automate · AI Agent Systems · Energy Policy

Power AutomateMicrosoft Azure

An executive newsfeed that curates itself.

8 hours → 5 hours a week. Every Monday, hands-off.

A four-flow Power Automate system that ingests newsletters and forwarded links from an inbox, resolves tracking URLs to real articles, uses Azure OpenAI to triage and summarize, lets the analyst approve what matters, then generates a fully-formatted weekly executive briefing with clickable hyperlinks.

Email-triggered intake. Every forwarded link or newsletter becomes a SharePoint row, auto-deduped.
Dual-path extraction. Long newsletters get AI-extracted; single links run through a resolver chain.
AI triage. Azure OpenAI writes a 3-sentence summary, takeaways, and a why-it-matters line.
One-click report. Approved articles render into a branded Word doc with live hyperlinks.

Weekly Energy Brief

Monday · 07:00

Policy

Markets

Innovation

M&A

Case study

An executive newsfeed that curates itself.

8 hours → 5 hours a week. Every Monday, hands-off.

The problem

An energy policy consultancy's principals spent a full day every week curating a weekly news briefing for executive clients. The analyst triaged 60–80 forwarded links and newsletters, clicked through Mailchimp and tracking wrappers to find real articles, wrote a summary for each, sorted them into Ontario / Canada / USA / International across climate, electricity, nuclear, and biomass subcategories, then manually built a Word document with clickable hyperlinks. Coverage was inconsistent, junk links cluttered the list, and the formatting step alone took two hours.

The solution

Four decoupled Power Automate flows that isolate ingestion, triage, approval, and report generation. An Office Script extracts URLs from raw email HTML. An HTTP resolver chain unwraps tracking links, decodes base64-wrapped redirects, and pulls canonical URLs, page titles, and Open Graph metadata from the destination. Azure OpenAI triages each article into a summary, takeaways, and a why-it-matters line. An Azure Function running docx-js renders approved articles into a fully-formatted Word briefing with live hyperlinks — on demand, from a single button click.

How it flows

Ingest

Email trigger
Office Script URL extract
Redirect resolver chain

Triage

Azure OpenAI summary
Takeaways + why it matters
Geography suggestion

Review

SharePoint list
Status: New / Approved / Junk
Flag top issues

Render

Azure Function, docx-js
Hierarchical sections
Live hyperlinks

Steps to solve

01
Map the real intake. The analyst's "newsfeed" was actually two distinct input types — single-link forwards and dense multi-article newsletters — each needing a different extraction path.
02
Build the URL resolver. Office Script parses href attributes with string ops (no regex — runtime doesn't support it). Power Automate HTTP follows redirects; Compose steps check for canonical tags, og:url, base64-encoded tracking wrappers, and JSON-LD structured data.
03
Handle the newsletters. Long emails branch to an AI extraction path that pulls headlines, URLs, and summaries directly from the email body — skipping the per-article fetch entirely.
04
Filter the junk. Ruleset catches social links, newsletter furniture, app store links, hash-string headlines, and bare homepages. Anything ambiguous gets tagged Junk, not deleted — the reviewer decides.
05
Triage with AI. Flow 2 sends the article body to Azure OpenAI with a structured prompt; JSON response is parsed and written back to the row. Base64 wrapping was ruled out early — GPT hallucinates on encoded content.
06
Survive bot blocks. Reuters, Guardian, phys.org actively block automated requests. Every downstream step has both-success-and-failure run-after settings so the row still gets created with whatever metadata was recoverable.
07
Render the report. Azure Function on Consumption plan receives approved articles as JSON, maps choice fields to template sections, emits a docx with H1/H2/H3 hierarchy, a Top Issues bullet section, and clickable hyperlinks.

Inputs & context

News Intake mailbox. Dedicated Outlook folder rule; newsletters forwarded in or subscribed directly.
SharePoint list. 17 columns covering status, headline, link, publication, geography, topic, subtopic, AI fields, top-issue flag, and email metadata.
Azure OpenAI deployment. GPT-4o-mini in Canada East; triage and deep-analysis prompts tuned against the client's editorial voice.
Helper workbook. HelperScripts.xlsx on SharePoint holds the ExtractURLs Office Script — shared across all flows.
Editorial rubric. Geography order, topic-to-subtopic mapping, and article format (headline, publication, date, blurb) encoded into the report generator.
Word template. Client's existing briefing format reverse-engineered into docx-js heading styles, bullet numbering, and hyperlink runs.
Azure Function. Node.js 22 on Consumption plan, same resource group as OpenAI; receives article JSON, returns base64-encoded docx.

Outcome

Analyst time down from 8 hours to under 5 hours per week. Manual URL clean-up, summarization, and Word formatting replaced by a system that runs hands-off through the week and renders the final briefing on a single button click. Editorial judgment stays with the analyst; everything else is automated.

Visuals

↓

Screenshot placeholder

Power Automate flow view — four decoupled flows from email trigger to report render.

Screenshot placeholder

Rendered weekly briefing — hierarchical sections with live article hyperlinks.

2 more projects hiding below the fold.

Built, not just talked about.

70+

founders, executives, and board leaders advised — including leaders at Verizon, USAA, The Honest Company, and the Terry Fox Foundation

50+

custom agents, GPTs, tools, and knowledge systems built and deployed across live client workflows

45+

workshops, strategy sessions, and executive briefings across North America and the UK

30–50%

average time reduction for clients delivered through redesigned AI-assisted marketing, research, and operations workflows

Contact

If you got this far,
you already know what I do.

The question is what we should build together.

teryl@electrickitemedia.com

Talk soon, Sitemapxml.

Oh hey, Sitemapxml! So glad you're here!

Built systems, not just decks.

A site that knows who's walking in.

A custom AI system built around your context.

An Instagram engine built for a real small business.

A client portal that always knows which ads are running.

A grocery list that builds itself from weekly orders.

An executive newsfeed that curates itself.

Built, not just talked about.

If you got this far,you already know what I do.

If you got this far,
you already know what I do.