Every technology decision is a bet. Every platform choice is a statement about what the company thinks matters. You can read an earnings call and hear what management says they're building. Or you can read the job postings, the open source repos, the conference talks, and the vendor integrations — and see what they're actually building.
Here's what People Inc's tech stack tells us about where the engineering organization is placing its chips.
The Knowledge Graph Nobody Talks About
The most interesting thing in People Inc's stack isn't D/Cipher. It's Graphene-Selene.
Graphene-Selene is an internal knowledge graph and content services platform that powers the core APIs behind DDM's publishing infrastructure. It's built on GraphQL, Spring Framework, graph databases, and a microservices architecture running on GCP and AWS. The team is based primarily in Edmonton, Alberta — DDM's secondary engineering hub — and they're actively hiring, including interns who will "develop the next generation of the knowledge graph, content services, and the core APIs that drive our publishing and web platforms."
That quote is revealing. Graphene-Selene isn't a content management system — it's the semantic layer that sits between the CMS and the rendering layer. It understands what content means, not just where it lives. A knowledge graph with 40 brands' worth of content, structured as entities and relationships rather than flat documents, is the kind of infrastructure that makes everything else possible: cross-brand content discovery, D/Cipher's contextual targeting, taxonomy-driven ad placement, and eventually, the Inversion's product platform.
This is a build, not a buy. And that's the right call — no off-the-shelf knowledge graph understands the semantic structure of 40 media brands spanning food, health, finance, entertainment, and home. This is proprietary competitive advantage disguised as infrastructure.
The Speed Play
People Inc's rendering stack is stubbornly server-side. Java and Freemarker generate HTML on the server. Vue.js provides progressive enhancement — interactive components hydrate on top of server-rendered markup. SCSS handles styling through an atomic design component library shared across brands.
This is a deliberate architectural choice, not legacy inertia. Jon Roberts — DDM's Chief Innovation Officer since 2013 — has been explicit about the philosophy: "the fastest website, the best content, no bad ads." Server-side rendering means the browser gets complete HTML immediately. No JavaScript framework has to boot before the user sees content. Combined with a dual-CDN strategy (Fastly + Cloudflare), this is how you pass Core Web Vitals at scale across 40 properties when your competitors are shipping 3MB of ad tech JavaScript.
The tradeoff is real. Server-rendered Java templates are harder to iterate on than a modern React/Next.js stack. Recruiting Java developers for a media company is harder than recruiting TypeScript developers. But the choice optimizes for what matters most to DDM's business model: page load speed directly affects ad viewability, which directly affects CPMs, which directly affects the roughly $1 billion in annual digital revenue. When your ad strategy is "fewer, better ads at higher prices," every 100 milliseconds of load time is money.
The Data Spine
The data architecture reveals where DDM actually spends its engineering calories. BigQuery is the primary analytical warehouse, with Snowflake as a secondary environment. Kafka handles event streaming. Spark and Beam process batch and streaming workloads. Looker provides business intelligence.
But the most telling piece is Switchboard — a data platform founded by the team that launched Google BigQuery. DDM's four-person Revenue Analytics team uses it to process two billion rows of daily data across 40 brands and 20 partners without writing ETL code. When the Time Inc acquisition brought a 200% increase in data volume, Switchboard scaled without engineering intervention. DDM credits it with saving 5,000 engineering hours per project.
This is the composable data stack pattern: BigQuery for storage and compute, Kafka for real-time, Switchboard for transformation and integration, Looker for visualization. No proprietary data warehouse. No custom ETL pipelines. The engineering effort goes into what's unique — the semantic layer in Graphene-Selene and the targeting intelligence in D/Cipher — while commodity data plumbing is outsourced to best-of-breed vendors.
The Marketing Stack
The activation layer is vendor-composed: Keycloak for authentication, Hightouch as the composable CDP, Iterable for email/messaging, Piano for paywalls and subscription management, LiveRamp for identity resolution, Salesforce for CRM, Pushly for push notifications, and Digioh for on-site engagement.
This is the "buy everything except the brain" strategy. No custom CDP. No custom email platform. No custom identity graph for marketing (D/Cipher has its own for ad targeting). Every component is replaceable. The bet is that competitive advantage in marketing technology comes from the data and intelligence feeding these tools — Graphene-Selene's semantic understanding and D/Cipher's intent signals — not from the tools themselves.
The risk: integration complexity. Eight marketing vendors means eight APIs, eight data contracts, eight vendor roadmaps to track. A four-person analytics team managing this across 40 brands is tight. Hightouch as the CDP layer helps — it's the router that moves audience segments from the data warehouse into activation tools — but this architecture is one vendor sunset away from a scramble.
What's Missing
What you don't see in the stack tells you as much as what you do.
No ML platform. No references to MLflow, SageMaker, Vertex AI, or any model training infrastructure. D/Cipher uses OpenAI's models. But there's no evidence of internal ML engineering for content recommendation, personalization, or predictive analytics. This is a gap if the Inversion strategy requires personalized product experiences at scale.
No real-time personalization engine. The stack is optimized for fast delivery of the same content to every visitor. Server-side rendering plus aggressive CDN caching is fundamentally at odds with per-user personalization. To personalize, you either break the cache (destroying the speed advantage) or personalize at the edge (which requires infrastructure DDM hasn't publicly invested in).
No headless CMS. Graphene-Selene provides a GraphQL API layer, but the underlying content management still appears to be the proprietary Dotdash CMS that was migrated onto all 40 brands post-merger. True headless — where content is completely decoupled from presentation — would enable the Inversion's product experiences. Graphene-Selene may be the bridge, but it's serving the current publishing model, not a product platform.
Reading the Competitive Stack
| Capability | People Inc | Future plc | Condé Nast | Vox Media |
|---|---|---|---|---|
| CMS | Proprietary + Graphene | Vanilla (proprietary) | VERSO + Copilot | WordPress VIP |
| Ad Tech | D/Cipher (proprietary) | Hawk (proprietary) | 1P + standard stack | Concert (scaled back) |
| Data Platform | BigQuery + Switchboard | Aperture (proprietary) | Databricks | Standard |
| Knowledge Graph | Graphene-Selene | None public | None public | None public |
| ML/AI | OpenAI (external) | Limited | Databricks ML | Minimal |
| Build vs Buy | Build core, buy edges | Build everything | Build CMS, buy data | Retreated to buy |
The comparison reveals four distinct strategies for the same problem:
Future plc builds everything — Vanilla (CMS), Hawk (ad tech), Aperture (data). Fully proprietary. This gives them total control but requires massive engineering investment relative to revenue. When Google traffic collapsed, that fixed cost became a liability.
Condé Nast builds the editorial tools (VERSO, Copilot) but buys the data platform (Databricks). This gives them the strongest ML/AI foundation in the cohort — Databricks' lakehouse architecture is purpose-built for the kind of content-to-commerce analytics that the Inversion strategy requires. If DDM's biggest stack gap is ML infrastructure, Condé Nast's biggest advantage is having it.
Vox Media is the cautionary tale. They built Chorus, an ambitious proprietary CMS that was supposed to power the future of digital publishing. It didn't survive contact with economic reality. They retreated to WordPress VIP. The lesson: building a platform is expensive, and the window between "strategic investment" and "stranded cost" is narrow.
What the Stack Says
People Inc's technology stack makes three bets:
Bet one: speed over flexibility. Server-side rendering, aggressive caching, dual CDN, fewer ads. The entire architecture optimizes for the fastest possible page load. This is the right bet for an ad-supported publisher — but it constrains the Inversion. Product experiences (think MyRecipes planners, InStyle's The Intern, People App) will increasingly need dynamic, personalized, JavaScript-heavy interfaces that fight the speed-first architecture.
Bet two: semantic understanding is the moat. Graphene-Selene and D/Cipher together represent a decade of investment in understanding what content means, not just where it lives. The knowledge graph connects entities across 40 brands. D/Cipher turns that understanding into ad targeting. This is the right long-term bet — in a world where AI can generate content instantly, understanding and organizing content at scale is the defensible position.
Bet three: build the brain, buy the body. Proprietary knowledge graph. Proprietary ad intelligence. Everything else — data warehouse, CDP, email, auth, push, paywall — is a vendor. This keeps the 500-person engineering team focused on what's unique. But it also means People Inc's product differentiation runs through exactly two systems: Graphene-Selene and D/Cipher. If either falters, there's no proprietary fallback.
The stack is well-architected for 2024's problem: winning the ad game post-cookies with contextual intelligence and fast sites. The question is whether it's architected for 2027's problem: turning 19 media brands into product businesses. That requires personalization, ML, and dynamic experiences that the current stack was explicitly designed to avoid.
This analysis reconstructs People Inc's tech stack from job postings, vendor case studies, conference talks, open-source contributions, and industry reporting. It represents what's publicly observable — the actual internal architecture may differ. Sources include Graphene-Selene job listings, Switchboard case study, Digiday interview with Jon Roberts, and Digiday's merger integration coverage.