PulseMedia
Generative AI Content Engine
How we built a custom RAG-based content engine for a digital marketing agency that reduced first-draft production time by 78% while preserving each client's unique brand voice.
Content Output
3.1x more per writer
Draft Speed
78% faster first drafts
Project Overview
PulseMedia is a 25-person digital marketing agency managing content production for 40+ B2B clients across SaaS, fintech, and professional services. Their core challenge: each client has a distinctly different brand voice, messaging framework, and audience—meaning they could not simply adopt a generic AI writing tool without producing generic, off-brand content.
Intellectual Clouds was engaged to architect and build a custom Retrieval-Augmented Generation (RAG) content engine. Each client in the system has their own private knowledge base—containing brand guidelines, past high-performing articles, target persona definitions, and style notes—that the AI retrieves from before generating any content.
Client
PulseMedia Agency
Industry
Digital Marketing / Content
Timeline
10 Weeks
Services
Generative AI Development, RAG Architecture, Custom AI Solutions
The Challenges
Brand Voice Inconsistency at Scale
PulseMedia's writers were juggling 40+ clients daily. With growing demand, maintaining each client's distinct voice across all content was becoming impossible.
Off-the-Shelf AI Tools Produced Generic Content
Tests with ChatGPT and Jasper AI produced content that read as generic and failed client QA reviews 60% of the time, requiring near-complete rewrites.
Client Knowledge Siloed in PDFs and Emails
Each client's brand guidelines, personas, and strategic context were scattered across folders, email threads, and spreadsheets—inaccessible to AI tools.
Freelancer Cost Pressure
To meet growing demand, PulseMedia was spending $280K/year on freelance writers. Leadership needed a way to scale content without proportionally scaling headcount.
Before AI Engine
60% QA Fail Rate
on generic AI drafts
Our Solution
We built a multi-tenant RAG content platform where each client has an isolated, private vector database containing their brand knowledge. When a writer initiates a content request, the system retrieves the most relevant chunks from that client's knowledge base before prompting the LLM—ensuring every output is grounded in their specific brand context.
Per-Client Vector Knowledge Bases
Each client's brand guidelines, audience personas, past articles, and style notes were chunked, embedded, and stored in a dedicated Pinecone vector database namespace.
Structured Prompt Templates
Writers fill in a brief structured form (topic, angle, CTA, audience) rather than writing a prompt from scratch. The system assembles the final prompt by injecting the retrieved brand context.
Multi-Model Output Pipeline
First drafts are generated via GPT-4o for quality. A second pass with a fine-tuned headline model optimizes titles for AEO. A final SEO scoring pass validates keyword density before delivery.
Writer Dashboard & QA Workflow
We built a custom Next.js dashboard where writers generate, review, edit, and approve AI drafts. All outputs are tracked, enabling PulseMedia to continuously improve their brand knowledge bases.
Technical Stack
OpenAI GPT-4o (primary) + fine-tuned headline model
Pinecone (multi-tenant namespaces per client)
OpenAI text-embedding-3-large
LangChain (retrieval pipeline + prompt chaining)
Next.js 14 + Tailwind CSS (writer dashboard)
AWS Lambda + S3 (document ingestion pipeline)
The Results
Within 60 days of full deployment, PulseMedia's content team was producing 3.1x more publishable articles per writer per week. The QA fail rate on AI drafts dropped from 60% to under 8%. And for the first time, new client onboarding took hours (loading their brand docs into the knowledge base) rather than weeks (training a new freelancer).
Reduction in first-draft production time
More publishable content per writer
QA fail rate (down from 60%)
Annual freelance cost eliminated
We tried three off-the-shelf AI writing tools before this. They all produced the same thing: generic, forgettable content that our clients rejected. What Intellectual Clouds built is fundamentally different—it actually understands our clients because it reads their brand documents first. Our writers use it every day and our client retention has gone up because the content quality has gone up.
Sarah Whitfield
Managing Director, PulseMedia Agency
Ready to Scale Your Content Without Scaling Headcount?
Let's build a Generative AI content engine that preserves your brand voice while dramatically increasing output.