About Client
Workforce Group is one of Africa’s leading HR and business consulting firms. They help organisations
across the continent with end-to-end human resource solutions — from hiring and talent management
to leadership development and HR outsourcing. They operate at a scale comparable to Big 4 consulting
firms in the African market
What They Do
- Business consulting and HR strategy
- Recruitment and talent management
- Employee training and leadership development
- HR outsourcing and technology solutions
Their Clients Include
GTBank, Nestle, Access Bank, Dangote, Lafarge, and many other top companies across Financial
Services, Energy, Manufacturing, IT, and Public Sector industries
Problem Statement
Workforce Group had hundreds of PowerPoint presentations containing years of consulting knowledge
— service descriptions, case studies, best practices, industry insights. All of it was scattered across
Google Drive with no way to search, retrieve, or reuse it efficiently. A goldmine of institutional
knowledge, locked inside files nobody could find.
The Challenge
Over the years, the company built a library of 450+ presentations. Every client engagement, training
programme, and strategy session produced new slides. But this knowledge was never indexed, never
searchable, and never connected. When someone needed information, they opened folder after folder,
file after file — hoping to find something relevant.
What Was Going Wrong
1. Wasted Time on Search and Content Creation
Manually searching through dozens of PPT files took 30–45 minutes per search — and still missed the mark. Creating a single blog post or presentation took 4–5 hours. The marketing team spent more time searching for information than creating with it.
2. Repetitive Work
Team members spent hours copying content from old presentations and rewriting it for new purposes. The same insights were being reformatted again and again, by different people, across different teams.
3. Inconsistent Quality
Without a single source of truth, external communication varied in tone, structure, and accuracy across every deliverable. Different people wrote content in completely different styles.
4. Knowledge Waste
Valuable insights buried in old presentations were rarely reused. The knowledge reuse rate was just 12% — meaning 88% of what the team had already built was effectively invisible.
Solution Provided
We built an AI-powered system using RAG (Retrieval-Augmented Generation) architecture,
orchestrated through N8N workflow automation. The system reads every presentation Workforce
Group has ever created, understands the content inside, and lets any team member ask questions or
generate new content in seconds — with source citations from the original documents.
We chose Qdrant for vector storage because it supports hybrid search — combining keyword matching
with semantic similarity in a single query. Built and deployed in 6 weeks.
What is RAG in Simple Terms?
Think of it like a research assistant who has read every document your company has ever produced.
When you ask a question, they don’t guess — they find the exact slides, paragraphs, and data points
that answer it, then write a polished response with references to where each fact came from.
Traditional search matches keywords. RAG understands meaning. It knows that “employee retention
strategies” and “how to reduce staff turnover” are asking the same thing — even if the words are
completely different.
How the System Works:
01
Automatic Knowledge Collection
- Monitors Google Drive for new or updated presentations automatically
- Detects uploaded files, validates format, and queues for processing
- Knowledge base stays current — no manual intervention needed
02
Smart Processing
- Converts presentations to PDF for vision-based extraction via Mistral AI
- Extracts all text, tables, and image content from complex slide layouts
- Breaks content into semantic chunks and stores as vector embeddings in Qdrant
03
Intelligent Search & Retrieval
- Searches 78,000+ indexed knowledge chunks instantly
- Hybrid search — BM25 keyword matching combined with semantic meaning
- 91% relevance score across all queries
04
AI-Powered Content Generation
- Retrieved knowledge fed into OpenAI model to generate brand-aligned content
- Produces blog posts, presentations, executive summaries, proposals, and Q&A responses
- Every output includes source citations — grounded in company knowledge, not generic AI
Results
System Performance
| Metric | Value |
|---|---|
| Content Creation Speed | 75% faster — 4–5 hrs down to under 1 hr |
| Search Time | 30 min → 20 seconds |
| Knowledge Reuse Rate | 12% → 73% (6x increase) |
| Manual Hours Saved | 320 per month |
| Blog Posts Generated | 180+ in first 6 months |
| Knowledge Chunks Indexed | 78,000+ from 450+ presentations |
Operational Impact
- Content team capacity increased 3.5x — without hiring additional staff
- Client proposal creation time reduced by 70%
- First-time content approval rate up from 45% to 82%
- Content production cost reduced by 65%
- Employee onboarding time cut by 40%
Before vs After
| Dimension | Before | After |
|---|---|---|
| Search | Manual folder browsing, 30–45 min | Instant semantic search, 15–20 seconds |
| Content Creation | 4–5 hours per blog post | Under 1 hour per blog post |
| Knowledge Reuse | 12% of existing content | 73% reuse rate |
| Content Approval | 45% first-time approval | 82% first-time approval |
| Team Capacity | Limited by headcount | 3.5x capacity, no new hires |
| Knowledge Access | Locked in scattered PPT files | Searchable, connected, always current |
Tech Stack
| Layer | Technology | Role |
|---|---|---|
| Orchestration | N8N | Workflow automation & pipeline management |
| Vector Database | Qdrant (self-hosted) | Semantic storage, hybrid search & retrieval |
| Source Storage | Google Drive | Presentation files & document management |
| Document Understanding | Mistral AI | Vision-based text extraction from documents |
| AI Model | OpenAI ChatGPT | Content generation and Q&A |
| Embeddings | Cohere | Text-to-vector conversion |
| Search | Hybrid (BM25 + Semantic) | Keyword + meaning-based retrieval |