QueerTech 2023–2024 · Caitlin Lee Consulting 2022–2023
8,000 professionals, 50% engagement growth
I built engagement systems for professional communities at two scales: 50% social engagement growth for an 8,000+ tech network, and learning infrastructure serving 25+ social justice organizations.
one node becomes a network
A network to serve, not to broadcast at. Engagement up 50%, and a resource centre so members could find what we made.
Two roles, one skill
Community work shows up on my resume twice, in different shapes. At QueerTech I managed digital marketing and content for a large professional network. At Caitlin Lee Consulting I built client network and learning programs for a portfolio of mission-driven organizations. The underlying skill is the same: making a group of people more useful to each other.
At QueerTech
Digital marketing manager for a network of 8,000+ 2SLGBTQ+ tech professionals, growing social engagement 50% during my tenure. The work was making the network valuable enough that members took part rather than passively received. I built a centralized resource centre on Hivebrite so members could reach podcast episodes, event recordings, career resources and sponsor content on their own, which lightened the load on staff and gave members more. The podcast, the conference coverage, the sponsor integrations and the member spotlights were all designed to make the network more useful to its members, not just more visible to outsiders.
At Caitlin Lee Consulting
Led client network programs for a consultancy serving 25+ social justice, direct action and direct service organizations, from grassroots collectives to established non-profits. Built scalable learning programs, including curriculum design, video courses and skill-tracking tools, all of which had to work for organizations with very different capacities and contexts. And I treated the client network itself as something to design. How do organizations with similar challenges find each other, and how do insights from one engagement benefit the whole portfolio?
Honest scope note. At QueerTech I was one person on a small team; the community's growth was a team effort, not a solo achievement. At Caitlin Lee I led the network and learning function but worked within a consultancy model. I built systems; I did not build either organization.
What this demonstrates
Community operations is not social media management. It's designing the systems that make a group of people more valuable to each other: resource access, connection pathways, learning infrastructure, and engagement loops that sustain themselves. The same discipline that grew a tech professional network built learning systems for social justice organizations.
Artifacts on request: resource centre structure, engagement metrics summary, learning library architecture, curriculum samples (sanitized).
Want the artifacts?
Sanitized versions of the specs, rubrics and inventories behind any of these are available on request. Tell me which lane you're hiring for and I'll send what's relevant rather than everything.
Each one names the problem precisely, shows what I built, and says plainly what I cannot claim. Where a program was designed but not fully deployed, it is labelled that way rather than counted as an outcome.
Premier Health of America · content systems · 2024–2026
From four-plus weeks to five days: what an AI-native content pipeline actually compresses
Producing the first on-brand article for a new organization used to take four-plus weeks, and most of that was not writing. It was finding the right subject-matter expert, and reverse-engineering a brand's voice from everything it had already published. The system I built took that down to five days, piloted as a full pipeline: Monday to Friday, request to published.
the cascade, resolving
corebaseline voice, non-negotiable values
brandonly how this brand departs
surfaceblog, release note, microcopy
aspectscompliance, terminology, audience
↓
one promptfully resolved, same manifest in, same prompt out
The compiler makes no model calls. Determinism where reliability is required, judgment where originality is.
The speed is the least interesting thing about it. The system doesn't produce articles; it produces prompts. That distinction is the whole story, because it is what separates a durable content system from a faster way to generate a pile of drafts.
The two costs nobody counts
Writing the first piece for a new organization is slow for two reasons that have nothing to do with typing speed.
The first is relationship cost. The biggest time cost when I enter a new organization is finding the subject-matter experts, right down to what their email is, whether I have a relationship with them, or whether I know someone who can get me to them. None of that is writing. It is locating the right person, then finding a credible way into their time.
The second is brand-voice cost, and it's the less obvious one. Writing convincingly in an established voice means auditing everything previously published to reverse-engineer that voice, then writing to match it and improve on it. It's easy to write like yourself. It is not easy to write like someone else. Multiply that across several distinct brands and the cost compounds, because each switch is less like changing a style guide and more like learning the voice from scratch.
An LLM doesn't fix either one
A model doesn't know who the right expert is, and it doesn't know a brand's voice any better than a new hire does on day one. It will write in a confident, generic register unless something constrains it. Applying that constraint by hand, for every piece and every brand, is exactly the re-learning loop above. The technology alone doesn't fix the bottleneck; it just makes the first wrong-voiced draft arrive faster. That is how a team ends up with a pile of AI drafts that all sound like nobody in particular.
The real problem was never whether a model can write. It was whether voice, compliance and audience can be applied reliably, the same way, every time, without a person re-deriving them from scratch on each piece. That isn't a writing problem. It's a systems problem, and it wants a systems answer.
The compiler
The shift was architectural, not tooling. The system treats content the way a compiler treats code: brand voice, compliance floors, terminology and audience guidance are each defined once, as versioned layers, then compiled deterministically into a single working prompt for a given task. You don't teach a writer, or a model, a brand's voice fresh for every piece. You write one general-purpose draft, and the system applies the brand as a layer on top.
The stack resolves general to specific
Core. An organization-wide baseline voice plus a small set of non-negotiable values, inherited by every brand and changed rarely.
Brand. A cascade from owner brand through operating brands to sub-brands, each declaring only how it departs from the baseline.
Surface. The format rules for what is being written: blog post, press release, product microcopy, release note.
Aspects. Cross-cutting layers for compliance, controlled terminology and audience, attached automatically by the task rather than remembered by a person.
A manifest names the task: brand, surface, audience, and a short brief. The compiler walks the stack and stamps out one prompt with everything already resolved, and it makes no calls to a language model itself. It is deterministic, so the same manifest produces the same prompt every time. The creativity lives in the human-plus-model step that drafts against the compiled prompt; the reproducibility lives in the compiler. Putting determinism where you want reliability, and human judgment where you want originality, is the most important design decision in the whole thing.
Voice is a diff; compliance is a ratchet
A brand doesn't restate its whole voice. It declares a small set of bypasses against the baseline: tone, framing, lexicon. That is why "write once, translate into voice" is architecturally true rather than a slogan. Applying a brand isn't a rewrite; it's a diff layered onto a shared foundation.
And not everything is bypassable. Compliance and controlled terminology are floors, not preferences: they may only ratchet stricter as the cascade descends. A sub-brand can sound more playful than its parent; it cannot become less careful. Strictest wins.
Review gates, and why nothing publishes itself
Four gates sit in the cycle rather than after it: a preflight scan for unresolved authoring markers and open decisions; a 24-category legal risk scan for a regulated industry; a findability scan; and stakeholder sign-off. Agents may propose research, but every fact carries a provenance tier checked against a registry of who is allowed to vouch for what, and an agent can never be recorded as a human source. Every status change is a person deciding, on purpose, with the labels in front of them.
Honest status. Shipped and in daily use by a non-technical team. Five days is the piloted number, not the theoretical floor. The pipeline compresses further only when every stakeholder is available at once, and full stakeholder availability never happens, for good reasons: the people whose sign-off the work needs are busy doing their own jobs. Adoption was uneven across functions, so the system was deliberately designed with guarantees that don't depend on how fast anyone picks up the tooling. The company entered creditor protection and was acquired in 2026. The architecture was built portable and has since been generalized brand-independent.
What this demonstrates
The interesting move isn't speed, it's where the reliability lives. Compile the constraints deterministically, then generate against them. A team that does the opposite gets faster output and no more trust than it started with.
Artifacts on request: layer schema, manifest examples, the 18-check rules engine inventory, the 24-category legal risk scanner, gate definitions (sanitized).
Premier Health of America · findability · 2025–2026
Making content findable to humans and AI
I designed a findability governance system that treats Google, AI assistants and on-site search as one problem, so a multi-brand healthcare staffing company could be found, cited and trusted in both languages. The company was acquired before rollout completed, so the KPIs here are the measurement system I built, not achieved results.
one question, three surfaces
searchsourced answer
ai assistantsourced answer
help centresourced answer
Same person, same worry, three doors. Grounding assets are designed to make all three say the same thing.
Context
The company operates multiple staffing agency brands with public websites, a career portal, an app help centre and a knowledge base. Content lived everywhere, was inconsistently structured, and was invisible in the places candidates actually look: search engines, job aggregators, and increasingly AI assistants.
Two concrete failures made the cost visible. Job postings across three agency sites did not surface in Google for Jobs or Indeed, because the applicant-tracking portals rendered in JavaScript. And four of the five analytics properties were dark, so nobody could measure what content did work. Neither was anyone's mistake. Both are what accumulates when a group grows by acquisition and inherits its tooling faster than anyone can govern it.
The problem, precisely
Findability was being treated as an SEO checklist. But the question had changed: when a nurse asks an assistant how to submit a timesheet, or which agencies staff northern Quebec, does the answer come from us, correctly, with a citation? No SEO checklist answers that. It takes content built for retrieval: chunked, anchored, disambiguated and grounded in verifiable claims.
What I built
One program specification covering six dimensions, written so a writer, a subject-matter expert, or an AI tool all receive the same compiled rules.
Audience tiering. Every document classified T1 public through T5 leadership-only, which decides where it lives and what search surfaces may see it.
A weighted scoring rubric. Technical SEO 25%, content quality 25%, authority and E-E-A-T 20%, AI retrieval readiness 20%, multilingual parity 10%. Pages get a number, not an opinion.
Grounding assets for retrieval. Each page ships with a machine-readable companion: what the page definitively covers, claims anchored to sources with verification dates, FAQ pairs, a glossary with synonyms, a disambiguation block naming what the page is not about, and guardrails defining when an assistant should refuse and where it should fall back.
Bilingual parity as a gate, not a hope. French-primary handling for the Quebec brand, Quebec terminology, gender-neutral phrasing, and a publish-blocking rule when parity is missing, designed so quality does not depend on any individual's French, including mine.
Compliance integration. ISO 9001 document control, meaning an SOP registry, change logs, approver signatures and review cycles, built into the content workflow instead of bolted onto it.
The spec embeds five reusable prompts (meta generation, schema generation, parity QA, grounding-asset building, retrieval spot-checks) so the system operates at the speed of the tools rather than the speed of a checklist.
How success was defined
The program shipped with quantified guardrails rather than opinions: organic click-through uplift of 12%+ on updated pages within 90 days, assistant answerability@3 of 0.8 or higher against a golden question set, citation rate 0.9+, hallucination rate under 0.05, French-English parity 0.95+, and a seven-day publish SLA for new content. Alert rules were part of the spec too. An answerability drop of more than ten points triggers a grounding audit; parity below threshold blocks publish.
Honest status. Those are designed targets. The company entered creditor protection and was acquired before the program completed rollout, so I present the KPIs as the measurement system I built, not as achieved results. The supporting layer did ship: the 18-check rules engine, the legal risk scanner, the UTM convention, and the analytics remediation plan that surfaced the measurement problem in the first place.
What this demonstrates
Findability is a governance problem, not a checklist. The same discipline that makes content citable by an assistant, meaning anchored claims, disambiguation and verified sources, is what makes it trustworthy to a nurse deciding whether to apply. Build the system once, and every page, in both languages, inherits it.
Artifacts on request: full program spec, scoring rubric, grounding asset template, rules engine inventory, legal risk scanner categories, UTM convention, analytics remediation plan (sanitized).
Premier Health of America · knowledge architecture · 2024–2026
Knowledge architecture for seven brands and six audiences
I designed the access model, taxonomy and voice system that let one small content team serve seven agency brands, six audiences and two languages, without the quality depending on any single person.
five tiers, checked on purpose
T1 publicanyone, including candidates
T2 internalall staff, all brands
T3 roleschedulers, recruiters, clinical
T4 restrictednamed owners only
T5 leadershipa decision, never an accident
Being left out of a document should be something someone chose and can explain.
The problem, precisely
Three entangled failures. Access: no model for who should find what, in an industry where a payroll policy and a public FAQ cannot live at the same trust level. Naming: the platform's data model and the company's content disagreed about basic entities, so search failed and teams talked past each other. Voice: seven brands with one content team meant the choice was either bland uniformity or unmanageable drift.
The access tier model
Role-based collections with five sensitivity tiers, T1 public through T5 leadership-only, with the role tier split between agency staff and practitioners (five tiers, six audiences), and the last tier deliberately stored outside the knowledge base entirely. The design decision that mattered: exclusion is a feature. Deciding what does not belong in a system is governance, and the tier model made that decision explicit and repeatable instead of ad hoc. In most organizations, being left out of a document is an accident nobody meant. Here it is a decision someone made and can explain.
The bilingual taxonomy, 270+ terms
A controlled vocabulary for the platform data model, hierarchical from top concepts down to field-level terms, with English and French definitions. Grounded in research rather than preference: I authored a bilingual survey instrument that had schedulers rate data categories by decision usefulness, so the taxonomy reflected how people actually work. Entity naming decisions, like practitioner versus employee and resource versus personne-ressource, were made once, documented and enforced. That ended the quiet chaos of every team inventing its own words.
The voice system
A modular, cascading brand voice and terminology governance system: shared foundations at the top, brand-specific layers below, compiled into a single content package a writer or an AI tool receives before drafting. Piloted with a five-day brief-to-publish sprint framework with compliance and findability review built into days three and four instead of bolted on after. The success criterion I am proudest of: workflow repeatable without redesign.
The operating layer
An Asana governance model with a portfolio taxonomy, required fields, status definitions and quarterly housekeeping, so the work about the work followed the same discipline. Supporting rules: 18 automated editorial checks including originals like Quebec Localization, Nurse Sensitivity and De-Patronize, plus a 24-category legal risk scanner for pre-publication review in a regulated industry.
Honest status. The tier policy, taxonomy, voice system, rules engine and Asana model all shipped and were in daily use. The company entered creditor protection and was acquired in 2026. The systems were built to survive handover, with versioned documentation and decision records. The archive of that work is what this portfolio is built from.
What this demonstrates
Knowledge architecture is deciding, once and explicitly, what things are called, who may see them, and how they should sound, then building the machinery that holds those decisions without a meeting. I design for the day I am not in the room.
Liphe · product design and information architecture
Information architecture before the first mockup
On Liphe, the nurse-placement platform, the team did the thing that almost never happens on a real product: defined the information architecture before anyone drew a screen. Alejandro led that work. I followed his lead and kept building into it: the data-model and workflow layers, the design system consolidated in one place, and the content structures on top.
structure first, screens after
Alejandro led this. Data model, then workflow, then content structures, and the screen inherits all three.
The problem, precisely
Products are almost never sequenced architecture-first. They are sequenced by urgency: give us a mockup and we'll build the front end. That pressure skips the structural work of deciding how information is organized before anyone starts drawing, and the architecture gets backfilled later, or never.
Backfilled IA is expensive in a specific way. The screens get built around whatever the first mockup implied, the data model accretes to match, and by the time anyone names the entities properly, renaming them means touching everything. On a platform where the same concept was already called different things in the product and in the content, that was a real cost, and a near-term one.
Architecture before interface
Under Alejandro's lead, the team defined a clear, bare-bones information architecture before front-end work began: what the entities are, what they are called, how they relate, and which surfaces they appear on. Naming decisions were made once and documented, grounded in research rather than preference. I authored a bilingual survey instrument that had schedulers rate data categories by how useful they were to an actual decision, so the structure reflected how the people using it work.
This is foundational in the way that is invisible while it is happening. It does not look like progress. It looks like progress later, when everything built on top of it moves faster than it otherwise would have.
Consolidating the design system
The decision that the design system would live in one place, single source, met real resistance. The concern was reasonable: that centralizing everything would produce confusion rather than clarity, and that the overhead wouldn't pay for itself.
That argument wasn't won by arguing. I moved fast and visibly through the data-model and workflow layers until the work itself made the case, and the people who had doubts came around having watched it work rather than having been told it would. That's worth naming as a technique, because it generalizes: on questions of tooling and structure, a demonstration beats a position paper almost every time.
It also had a cost. Pushing into the scheduling data model, which was another person's territory, while carrying design authority read, fairly, as stepping on toes. And it moved that work forward faster than it otherwise would have. Friction and forward motion were the same event, and I would rather name that plainly than pretend the win was frictionless.
Structure and language are the same project
The IA and the controlled vocabulary are one project seen from two angles. The 270-plus term bilingual vocabulary governing the platform data model is what keeps the architecture from drifting once product and content teams start moving independently, with entity naming enforced across both.
Two honest caveats. First, there is no isolated metric for the speedup. The pace improvement was real and visible across the team, but nobody ran a controlled comparison and I won't manufacture a percentage for one. Second, and more important: the conditions that made this sequencing possible were partly luck. The engineering culture on this product was unusually low-urgency, which created room most projects don't get. The transferable part isn't "convince leadership to slow down." It's recognizing foundational work when the room appears, and being willing to spend social capital to push it through.
What this demonstrates
Information architecture is a content problem before it is a design problem. Deciding what things are called, and how they relate, is the decision every screen inherits. Better to make it deliberately and early than to inherit it from whoever drew the first mockup.
Artifacts on request: the IA structure, taxonomy excerpt, scheduler survey instrument, design-system organization (sanitized).
Liphe · product content design · 2024–2026
Writing a consent threshold
A visibility toggle in our profile builder was quietly exposing nurse data before anyone had agreed to it. I rewrote it as a consent threshold, and the interface started telling the truth.
a consent threshold, not a setting
Make my profile visible to clients
Your credentials are being verified. This takes two to six weeks. Your profile stays hidden until it clears.
The old copy implied instant approval. The wait was always real; only the sentence changed.
Context
The platform matches nurses to agency placements. To be placed, a nurse builds a profile: credentials, availability, specialties, work history. Client agencies browse those profiles. Between a nurse filling in the form and a client seeing it there is a verification step, and provincial credential verification takes two to six weeks. That gap is where the problem lived.
The problem, precisely
The profile builder had a visibility control that looked like every other settings toggle in every other app: a switch, a label, a neutral default. But there was a path through the flow where profile data became accessible to client agencies before the nurse had explicitly turned anything on.
Framed as a settings control, that reads like a preference someone forgot to set. Framed correctly, it is a consent problem: personal professional data, in a regulated healthcare context, exposed without an affirmative act by the person it belongs to. That distinction is not cosmetic in a Quebec privacy context. I raised it as a product and consent question rather than a legal opinion, and flagged it for the people whose call that was.
The copy made it worse rather than better. A neutral toggle label implies the stakes are low. Nothing in the flow told a nurse what "visible" actually meant, to whom, or when it would take effect given their credentials might still be weeks from verified.
Renaming the problem
I argued the toggle was not a setting but a threshold: the moment personal data crosses from private to visible. Settings can have defaults. Thresholds require an act. Once the team accepted that framing, the design questions answered themselves, because you cannot default someone across a threshold on their behalf.
The redesigned copy and sequence made three things explicit at the moment of decision: what specifically becomes visible, who can see it, and how verification timing interacts with it, so a nurse isn't agreeing to something whose timing they can't see. Pre-toggle access was flagged as a consent problem, not a UX rough edge.
A standard, not a fix
The single screen was the smaller half. The larger half was writing the product content standard the rest of the app is held to: labels, navigation, forms, helper text, errors, empty states, onboarding, confirmations, notifications and operational prompts, across both the clinician app and the agency console. It opens with a hard hierarchy:
Clarity first, warmth second, personality last. A line fails if warmth, friendliness or brevity hides the state, action, consequence or next step.
That hierarchy is the whole argument in one rule. It's what lets a reviewer reject a line that is charming and unclear without it becoming a matter of taste.
Where it shows most: error states
Healthcare software is usually at its worst here. A representative rewrite:
Before: "Invalid credential." After: "Your BLS certificate expired on 28 May 2026. Upload a current certificate to apply."
Same event, same screen. The first version tells a nurse they are wrong. The second tells them what is true, when it happened, and what to do next. Multiply that across a credentialing flow used by tired people between shifts and it stops being a copy preference.
Honest status. The consent-threshold framing shipped in the profile builder, and the standard governs in-app content across both surfaces. I want to be straight about what I cannot claim: there was no before-and-after instrumentation on these flows, so I have no completion rate, no ticket-volume delta and no satisfaction score to show you. The measurement framework was designed and documented alongside the standard, but wasn't running long enough to produce a result before the acquisition. I'd rather hand you a defensible account of the reasoning than a number I didn't measure.
What this demonstrates
Product content design is not the sentence at the end of the process. It's being in the room early enough to notice a control is mislabelled in a way that carries legal stakes, having the vocabulary to name why, then writing the rule that stops the next twenty screens from repeating it.
Artifacts on request: the product content standard, the error and empty-state rewrite set, the consent-threshold flow rationale, the measurement framework (sanitized).
Shopify 2021–2022 · Premier Health of America 2024–2026
Recruitment marketing with receipts
I've done employer brand at both ends of the market: defining the EVP for a Canadian tech employer hiring in full public view, and fixing the unglamorous plumbing that decides whether a nurse in Manitoba ever sees your job post at all.
what the crawler reads
before
<div id="root"></div>
The listing was drawn by JavaScript after load. Google for Jobs and Indeed saw an empty container, so the role was never indexed.
after
Registered Nurse, Night Shift, Montreal
JobPosting · title · location · salary · date posted
Server-rendered and marked up, so it appears as a result.
I diagnosed this, then built an audit agent that checks every posting for it. A job nobody can find is a job nobody applies to.
Part one: Shopify, the strategy end
Joined the employer brand marketing team as an intern in July 2021; promoted to manager in about five months. Pivotal contributor to the Employer Value Proposition from kickoff to launch, a C-suite stakeholder project. Wrote paid campaign copy for talent audiences across EMEA, APAC and the Americas, including a Singapore campaign localized from audience research showing that market valued entrepreneurship, helpfulness and impact. Coordinated cross-functional, multi-national campaign teams. Grew the employer brand LinkedIn following 76% over my tenure. Led the Top Employer designation application.
Part two: healthcare staffing, the execution end
Healthcare staffing is recruitment marketing with the difficulty turned up: multiple agency brands, regulated claims, bilingual audiences, and candidates who are chronically burned out and justifiably sceptical.
An employer brand foundation. Five research-based nurse personas, from the overwhelmed early-career idealist to the isolated rural lifeline, each with motivations, frustrations and retention opportunities recruitment content could act on.
Multi-property campaigns across remarketing, search and regional ad accounts, including a Google Ads campaign that produced confirmed hires.
The plumbing fix. Diagnosed why job posts across three agency sites never appeared in Google for Jobs or Indeed: the applicant-tracking portals rendered in JavaScript, invisible to crawlers. Then built a biweekly AI audit agent posting structured summaries to Asana, so the problem stayed visible until it was fixed.
The measurement layer. UTM governance with a controlled vocabulary, and a content-to-hire attribution architecture designed to answer cost per hire by source.
I also rewrote the website copy across the company's agency brands, with the same claims compiled through the voice cascade, so seven properties stopped contradicting each other.
Honest status. The attribution architecture was designed; implementation was interrupted by the acquisition. The Shopify figures are from channel analytics during my tenure, and public recommendations from my team lead and a senior talent partner are on my LinkedIn profile.
What this demonstrates
Employer brand fails at the seams: between the EVP deck and the ad account, between the job post and the crawler, between the click and the hire report. I've worked every seam. Strategy that ignores plumbing produces beautiful campaigns nobody sees; plumbing without strategy fills the funnel with the wrong people.
Artifacts on request: EVP brief, Singapore campaign assets, persona set, UTM convention, attribution architecture summary (sanitized).
QueerTech · podcast production · 2023–2024
30 guests, 3 sponsors, 18% reach
I hosted and produced a tech podcast from nothing to Lightspeed's CEO, with Canada Media Fund, Google Chrome and RBC as sponsors, reaching 18% of the target professional audience.
thirty conversations, produced end to end
Dax Dasilva · LightspeedFlorencia Herra Vega · AlleyCorp NordDavid Cardon · Morgan Stanley
Booking a CEO is one skill. Getting them to say something worth keeping is a different one.
The problem, precisely
Podcasts fail in two places: they never ship, or they ship and nobody listens. The challenge was building a production system rigorous enough to land major sponsors and senior guests, while staying lean enough for a non-profit budget and a small team.
What I built
End-to-end production
Full producer scope: preproduction research and guest outreach, vocal coaching for hosts, intro and outro craft, music procurement, in-studio and on-site recording, training others in home recording setups, script review, and season strategy. Not just editing, but the whole arc from concept to publish.
Sponsor relationships
Secured Canada Media Fund, Google Chrome and RBC. That meant building a pitch that positioned the show as reaching an underserved professional audience with influence and disposable income, not as a charity case.
Guest booking at scale
30+ tech leaders from early career to C-suite, culminating in an interview with Dax Dasilva, CEO and founder of Lightspeed Commerce. Each booking required research, positioning and a reason to say yes beyond "please help our community." Booking a CEO is one skill; getting them to say something worth keeping is a different one.
Craft and distribution
Editing and mixing, meaning denoise, level, transitions and stingers, because the show needed to sound professional enough that sponsors felt their brand was protected and guests felt their time was respected. Then go-to-market strategy, captions and audiograms, podcast ad placements, partnership audits, and a media resource centre so community members could find any of it.
Outcomes
30+ guest conversations with tech leaders at every level
18% reach within an 8,000+ professional network
Sponsors: Canada Media Fund, Google Chrome, RBC
The show kept running after I left: a system, not heroics
Related experience
This wasn't my first production work: ten-plus years of freelance audio for dance, film and theatre, including the Musée des beaux-arts de Montréal, Judson Church and The Kitchen in New York, and Festivals Phenomena and Fringe. Co-producer of Madness: the Podcast, featured on Michael Moore's Rumble.
Honest measurement note. The 18% figure is target-audience reach within QueerTech's 8,000+ professional network, the audience the show was built for. It is not a claim about listenership beyond that network. The show link and episode archive are available to check it against.
What this demonstrates
Podcast production is project management in audio form: stakeholder alignment across sponsors, guests and hosts; quality control across audio, content and brand; and a system that ships consistently. I built it once, handed it off, and it kept running.
Artifacts on request: show link, episode archive, sponsor pitch deck, guest booking process, production workflow documentation (sanitized).
QT Qonference · event content production · 2023–2024
Six-person crew, 4× ROI
I directed a six-person media crew for a tech conference, delivering 4× ROI on marketing spend through content and branding strategy that extended the event's value beyond the room.
six feeds, one delivery
Ten years of AV direction underneath it, mostly the craft of knowing in advance what will go wrong.
The problem, precisely
Event content usually fails one of two ways: it's captured but never used, or it's used but feels like an afterthought. The goal was content production integrated with brand strategy from the start, so every hour of crew time generated assets the organization could use for months.
What I built
Crew direction
Directed six people on site, with video, photography and audio capture coordinated to maximize coverage without disrupting the event. Shot lists, positioning plans, real-time coordination.
Content and branding strategy
The 4× came from planning backward from how content would be used: social clips, sponsor deliverables, member communications, recruitment materials, archive footage. Every capture decision served a downstream purpose.
Production, not documentation
Framing, lighting awareness, audio quality and b-roll that told a story, which is the difference between "we recorded it" and "we produced it." Plus brand integration: visual identity, signage, stage design consultation, and making sure captured content matched brand standards. Sponsors needed to see their logos; members needed to see themselves.
Outcomes
4× ROI on marketing spend, measured as content value generated against production cost
A multi-month content pipeline from a single event
Sponsor deliverables completed; archive footage for future use
Honest scope note. I directed the media crew and owned content strategy. I did not manage event logistics, venue, catering or programming. My scope was content and brand, not operations.
Related experience
Ten-plus years of AV technical direction in performance venues: the Musée des beaux-arts de Montréal, Judson Church and The Kitchen in New York, Festivals Phenomena and Fringe. The conference was larger scale, but the discipline was the same: coordinating technical execution with creative intent, which is mostly the craft of knowing in advance what is going to go wrong.
What this demonstrates
Event content production is not videography. It's strategic capture: understanding what assets an organization needs, planning production to generate them, and directing a team to execute under live conditions.
Artifacts on request: event content samples, shot list template, ROI calculation methodology, crew coordination documentation (sanitized).
QueerTech 2023–2024 · Caitlin Lee Consulting 2022–2023
8,000 professionals, 50% engagement growth
I built engagement systems for professional communities at two scales: 50% social engagement growth for an 8,000+ tech network, and learning infrastructure serving 25+ social justice organizations.
one node becomes a network
A network to serve, not to broadcast at. Engagement up 50%, and a resource centre so members could find what we made.
Two roles, one skill
Community work shows up on my resume twice, in different shapes. At QueerTech I managed digital marketing and content for a large professional network. At Caitlin Lee Consulting I built client network and learning programs for a portfolio of mission-driven organizations. The underlying skill is the same: making a group of people more useful to each other.
At QueerTech
Digital marketing manager for a network of 8,000+ 2SLGBTQ+ tech professionals, growing social engagement 50% during my tenure. The work was making the network valuable enough that members took part rather than passively received. I built a centralized resource centre on Hivebrite so members could reach podcast episodes, event recordings, career resources and sponsor content on their own, which lightened the load on staff and gave members more. The podcast, the conference coverage, the sponsor integrations and the member spotlights were all designed to make the network more useful to its members, not just more visible to outsiders.
At Caitlin Lee Consulting
Led client network programs for a consultancy serving 25+ social justice, direct action and direct service organizations, from grassroots collectives to established non-profits. Built scalable learning programs, including curriculum design, video courses and skill-tracking tools, all of which had to work for organizations with very different capacities and contexts. And I treated the client network itself as something to design. How do organizations with similar challenges find each other, and how do insights from one engagement benefit the whole portfolio?
Honest scope note. At QueerTech I was one person on a small team; the community's growth was a team effort, not a solo achievement. At Caitlin Lee I led the network and learning function but worked within a consultancy model. I built systems; I did not build either organization.
What this demonstrates
Community operations is not social media management. It's designing the systems that make a group of people more valuable to each other: resource access, connection pathways, learning infrastructure, and engagement loops that sustain themselves. The same discipline that grew a tech professional network built learning systems for social justice organizations.
Artifacts on request: resource centre structure, engagement metrics summary, learning library architecture, curriculum samples (sanitized).
Want the artifacts?
Sanitized versions of the specs, rubrics and inventories behind any of these are available on request. Tell me which lane you're hiring for and I'll send what's relevant rather than everything.