The VectorMind product journey 2025
- CohesionX
- 26 August 2025
- 3:05 pm
- 0 comments
Pretoria, South Africa – August 2025
This article showcases how the core VectorMind AI products, Iris, Synaptic Search, Atlas, and our pioneering Knowledge Nodes solve complex business challenges by embedding Gen-AI at the heart of their design.
Each solution addresses a real operational problem: from automating the extraction of critical data in chaotic document environments, to enabling natural,
conversational access to business databases, to organising and surfacing insights from millions of enterprise documents and making expert knowledge instantly accessible in the field.
As compared to simple chatbots, VectorMind offerings utilise GenAI for deep understanding, thinking, and integration, translating unstructured data into reliable, actionable intelligence. Through the integration of business workflows with AI, our solutions help organisations minimise risk, boost productivity, and achieve the true potential of their information.
Knowledge Nodes: The first step toward domain intelligence
The Real-World Problem
In rural South Africa, farmers face a different kind of knowledge challenge. Decades of best practice, government guides, research, and advice are trapped in thick binders or scattered across PDFs. Most farmers have smartphones, but little time to search for answers, especially during crises or planting seasons.
The VectorMind Solution
The CohesionX team pioneered one of the first Retrieval-Augmented Generation (RAG) assistants for agriculture. Thousands of documents were embedded and indexed using GenAI models, and farmers could simply ask questions via WhatsApp or a mobile app, receiving tailored, context-rich responses based on deep domain content.
How Gen-AI Was Used:
- Embedded domain-specific documents for semantic retrieval
- Allowed free-form questions in almost any South African language
- Used Gen-AI models to generate grounded, accurate answers based on trusted content, not the open web
This was more than a chatbot; it was an expert-in-your-pocket, always on, always relevant.
Iris (Text-to-Schema): Extracting Order from Data Chaos
The Real-World Problem
Imagine the scene inside one of South Africa’s largest retailers’ food safety labs: a continuous stream of emails, each carrying a PDF of lab results from suppliers all over the country and beyond. No two labs use the same format. One embeds images, another prefers dense tables, and a third sends scanned forms with handwritten notes. Hundreds of results pile up every day, and a handful of quality assurance staff are tasked with making sure every item that lands on the shelves meets strict safety and quality thresholds.
But the only way to get data from those PDFs into the systems that matter was to open each file, hunt down the numbers, and copy-paste them into sprawling Excel templates one cell at a time. This process resulted in fatigue, introduced errors and delays, increased compliance risk, and affected the retailer’s reputation for quality. At this scale, even small mistakes mean big risks.
The VectorMind Solution
The team knew throwing more people at the problem wouldn’t help, and that basic OCR or template-based automation couldn’t handle the variability. So, we created something smarter from the ground up: Iris.
Iris is a Gen-AI-powered extraction engine designed to make sense of any document. It combines computer vision, advanced language models, and dynamic schema prompts to understand whatever you send its way: tables, scanned forms, images, PDFs, Word docs. It doesn’t just read its reasons.
How It Works:
- Iris detects the type of incoming file: text, image, or hybrid.
- If it’s a PDF or Word doc, advanced GenAI-powered extractors (like QuantumPDF and PythonDocX) break down the content, preserving tables, images, and layout.
- The content is passed through a GenAI engine that understands both the content and a dynamic “schema” describing what fields are important.
- Iris asks itself a series of “schema-driven questions” about the content (“What is the test result for Salmonella?” “What is the batch number?”), using natural language reasoning and vision to find and validate the right values, even if the document is structured unexpectedly.
- The answers are validated, structured as clean JSON, dropped into a database, and returned ready for dashboards, analytics, or compliance workflows.
Because Gen-AI powers the reasoning and extraction, Iris adapts to new formats with minimal reconfiguration. There’s no need to retrain the system for every lab, template, or new document layout.
Where the product is today
What began as a solution for food safety labs now powers data extraction for property groups, car dealerships, insurers, and more. Whenever there’s a flood of unstructured documents, Iris brings order, speed, and trust at the speed of business.
Synaptic Search (FlatQuery): Making Business Data Conversational
The Real-World Problem
Picture a car showroom, where sales staff understand customer needs. Customers won’t ask the sales staff, “Can you run a SQL query for sedans under R300,000 and with less than 50,000km?” Instead, customers want to ask in their own words: “What used automatics do you have for less than 300k?” However, online, the retailer’s search was confined to rigid boxes, dropdowns, checklists, and structured forms, which frustrated users and limited discovery.
As a result, online engagement lagged, inventory went unseen, and sales teams missed leads simply because the data was trapped behind a wall of database logic.
The VectorMind Solution
We wanted to open the gates, make searching for cars (or anything else) as easy as asking a colleague. So, we built Synaptic Search.
Synaptic Search is a GenAI-driven, natural language search system for structured business data. Rather than forcing users to adapt to the system, Synapse adapts to the user, translating human language into optimised, multi-layered database queries that enhance user experience.
How It Works:
- The user asks a question in natural language, via web, chat, or mobile: “Show me hybrid SUVs in Cape Town with sunroofs.”
- Synaptic parses the question, using GenAI (language models and semantic embeddings) to extract intent, entities, and constraints, even if the language is casual or ambiguous.
- For each relevant database field, Synapse combines:
- Embeddings compare the user’s intent to product descriptions, tags, and metadata.
- Fuzzy Matching: GenAI algorithms spot spelling mistakes or similar terms (“sunroof” vs. “panoramic roof”).
- Classic SQL Filtering: Hard filters for price, location, or specific specs.
- Weighted Scoring: All methods are scored and blended to surface the most relevant, not just the most literal, results.
- The system returns an ordered, relevant set of results, often with “Did you mean…?” suggestions, synonyms, and insights only possible when GenAI augments classic search.
This same approach powers internal tools too: staff find colleagues in payroll, book services, or find sales leads, all via natural language.
Where We Are Today
Synapse Search began in automotive retail, but now delivers conversational search across industries, retailers, payroll teams, lead gen systems, and more. Wherever business data needs to be discovered, Synapse puts GenAI at the heart of the experience, lowering barriers and driving engagement.
Atlas: Navigating the Enterprise Knowledge Ocean
The Real-World Problem
Consider the challenge faced by a top-tier legal firm with a digital estate sprawling across 430,000 SharePoint sites and more than 50 million documents. Lawyers and clerks spend hours just finding the right case, clause, or precedent. Metadata is inconsistent, document organisation is fluid, and time is always short. Traditional search brings back a haystack, not a needle and sometimes the most important answers live in unexpected places or subtle relationships.
For these professionals, speed and context are everything; knowledge delayed is value lost.
The VectorMind Solution
The team built Atlas, a Gen-AI-powered discovery platform, to transform this chaos into clarity. Atlas does more than search; it orchestrates discovery, mapping the relationships between documents, cases, and workspaces using graph theory and multiple, AI-powered search strategies.
How It Works:
- A user submits a question (“What case law supports section 13 exemptions for environmental compliance?”).
- Atlas consults its AI “planner” to determine which specialised search tools to use:
- Semantic search: Finds concepts and meaning using embeddings across document content.
- Metadata search: Uses Synapse Search to query tags, properties, and structured data.
- Keyword search: Finds exact matches or relevant phrases.
- External web search (if needed): Broadens context beyond the internal knowledge base.
- Atlas runs all relevant searches in parallel, aggregating results across millions of documents, then organises them as a graph connecting workspaces, documents, and even “chunks” (sections or paragraphs) by their relationships.
- A GenAI-powered scoring and ranking algorithm evaluates direct relevance, network context, and even “max nested scores” (if a document is relevant, so are its related sections), surfacing the highest-value answers and organising results by context, not just keywords.
- Atlas presents a structured, actionable response, often with direct quotes, links, and contextual snippets, allowing users to dive right in.
With Gen-AI, Atlas doesn’t just understand information; it coordinates, organises, and summarises it in a way that saves time and adds value.
Where the product is today
Atlas set a new standard in legal and is now becoming indispensable to brand companies, engineering firms, and anyone faced with mountains of unstructured data. Wherever context and relationships matter, Atlas, driven by GenAI, is there to make sense of it all.
Why Gen-AI is the fabric, not just a feature
What truly sets these products apart is how they move beyond the idea of AI as just a chatbot, making Gen-AI an essential thread woven through everyday business life. By integrating Gen-AI directly into the tools people rely on, it helps teams tackle challenges that used to be impossible with simple rules or manual effort. Gen-AI isn’t here to take jobs; it’s here to empower people, bringing extra intelligence and support to every task and decision. In this way, AI becomes a genuine partner, working alongside humans and elevating how businesses operate.
Ready to build your own assistants?
Mail us: info@cohesionx.co.za
About CohesionX
CohesionX is a South African technology company specialising Generative AI. Its flagship product, VectorMind, enables organisations to deploy AI-powered assistants that manage workflows, automate tasks, and drive intelligent decision-making; all within secure, compliant cloud environments.
Learn more at www.cohesionx.co.za and www.vectormind.online.
Media Contact:
Yaki Kruger, CohesionX
Email: yaki.kruger@cohesionx.co.za
082 841 4932
© Copyright 2023 - 2025 | CohesionX Pty Ltd | All Rights Reserved
Die Klubhuis | 1st Floor | Corner of 18th Str & Pinaster Ave | Hazelwood, Pretoria, Gauteng, 0010
