Atlas Architecture Center <> Partners Library: Purple Fabric

1. Enterprise Knowledge Garden: Data Transformation for AI

Purple Fabric's Enterprise Knowledge Garden (EKG) transforms structured, semi-structured, and unstructured data into decision-grade knowledge that is contextually rich, traceable, and continuously accessible by AI agents.

EKG ingests data from diverse sources and formats — including structured data from warehouses and APIs, semi-structured content like regulatory filings or industry reports, and unstructured assets such as PDFs, emails, and call transcripts. EKG uses a first-principles-driven approach to chunking and enrichment which mimics human understanding of content rather than relying solely on fixed token windows.

The role of MongoDB:

MongoDB Atlas serves as the data storage and retrieval backbone for EKG. Each enriched chunk is converted into a vector embedding and stored within MongoDB using native vector indexing. By leveraging MongoDB's flexible document model, embeddings are stored alongside rich metadata — including source, document type, timestamp, and data lineage — ensuring deep context and traceability.

This results in a dynamic knowledge repository that is not only searchable but intelligent, providing contextual precision when feeding insights into large language models (LLM). EKG supports tasks across various domains, such as summarizing a complaint case or cross-referencing contract clauses.

2. Enterprise Digital Experts: Multi-Agent Autonomous Systems

Purple Fabric's Enterprise Digital Experts (EDE) empowers users to build autonomous, domain-specific AI agent teams that can reason, collaborate, and take action independently. Agents are crafted using low-code orchestration and natural language prompts, operating as atomic, molecular, or compound entities. They leverage advanced frameworks, such as ReAct, to incorporate planning, memory, and inter-agent Socratic dialogue for more refined decision-making and outcomes.

The role of MongoDB:

MongoDB handles the dynamic memory and state management of these agents. Each agent maintains its own working memory, task state, and contextual knowledge within MongoDB, enabling real-time persistence and retrieval of evolving states during multi-turn reasoning or user interactions.

When factual grounding is needed, agents invoke Retrieval-Augmented Generation (RAG) pipelines to access the Enterprise Knowledge Garden (EKG). MongoDB Atlas powers this pipeline by delivering fast, filtered access to the right vector embeddings and source documents based on query intent and similarity. This ensures that agents provide responses grounded in traceable, contextually relevant enterprise knowledge which minimizes hallucinations and maximizes reliability.

3. Enterprise Governance: Operational Compliance and Control

Purple Fabric's Enterprise Governance (EG) stack enforces responsible AI by default, embedding compliance, security, and explainability into every layer of the system.

The governance layer oversees every aspect of the AI lifecycle, including data ingestion, agent orchestration, LLM invocation, and final output generation. Its design ensures that every action is observable, auditable, and policy-compliant, enabling enterprises to meet internal risk requirements and external regulatory standards.

Key governance capabilities include:

• Prompt governance and toxicity detection: Filtering unsafe or non-compliant queries and responses.

• Access controls and workspace isolation: Segmenting roles, teams, and visibility for secure collaboration.

• PII masking and entity redaction: Protecting sensitive data at both the retrieval and output layers.

• Rate limiting and usage tracking: Monitoring token usage, agent behavior, and preventing overuse.

The role of MongoDB:

MongoDB is foundational to Purple Fabric's governance framework, particularly in ensuring retrieval transparency and document lineage. When agents invoke RAG pipelines to ground their outputs, MongoDB stores embeddings, metadata, and source pointers for retrieved knowledge chunks. This creates a verifiable trail that maps the entire process — from input to retrieved knowledge to generated output — enabling post-hoc audits, compliance reviews, and root-cause analyses.

While Purple Fabric's control plane governs broader aspects like LLM safety, agent reasoning, and user workflows, MongoDB enhances traceability, security, and validation of knowledge access at scale, reinforcing the platform's ability to meet the demands of regulated industries with confidence.

4. Model Optimisation Hub: Dynamic Model Selection

Purple Fabric's Model Optimization Hub (MOH) is designed to give enterprises the freedom and intelligence to choose the right LLM for the task. It addresses one of the most critical challenges in scaling AI—balancing cost, accuracy, and performance across diverse use cases.

MOH allows teams to benchmark multiple LLM against real-world tasks such as summarisation, extraction, classification, reasoning, or validation. These evaluations are based on customisable metrics like latency, token consumption, output precision, and more, which are all tracked through real-time testing and usage data.

Once benchmarked, models can be:

• Pinned to specific agents or use cases based on performance

• Switched dynamically depending on cost/latency constraints

• Tuned with prompt engineering and few-shot examples for domain specificity

MOH also supports bring-your-own-model (BYOM) capabilities, allowing enterprises to plug in proprietary or fine-tuned LLM running on their own infrastructure or via private endpoints for use cases that demand higher control, data sovereignty, or custom domain adaptation. This modularity ensures that enterprises can avoid vendor lock-in, maintain performance across changing use cases, and optimize LLM economics without sacrificing quality or compliance.

Use Case: Complaints Investigation

Transforming a five-week manual process into an AI-driven solution that delivers results in minutes.

A leading wealth manager in the UK faced a crisis when new consumer duty regulations triggered a surge of over 50% in customer complaints. With more than 10,000 unresolved cases piling up, the firm was overwhelmed by lengthy investigation cycles. Solving a single complaint took up to five weeks, requiring investigators to manually pull information from over 10 systems—including policy admin databases, CRM tools, internal documents, and regulatory updates.

The process was burdened with inefficiencies: fragmented data sources, high error rates, and an inability to adapt to evolving policies and regulations. With less than 30% of complaints resolved within SLA timelines, the firm struggled with growing customer dissatisfaction and mounting compliance risks.

Purple Fabric transformed this process by deploying a production-grade multi-agent system for complaint investigation, designed to act, reason, and collaborate with speed and consistency.

The solution orchestrated a team of over 20 intelligent agents, each specialized for a part of the workflow. Some agents focused on extracting and logging complaint details, others verified policy terms, ran precedent checks, gathered evidence across systems, or wrote structured investigation reports. A Complaints Manager agent coordinated the entire flow, and a Human-in-the-Loop interface ensured compliance and oversight for final decisions.

The results?

What previously took five weeks of manual handoffs is now completed in minutes. The AI-powered system autonomously handles evidence gathering, classification, investigation, and adjudication with auditability and consistency:

• 22 processes integrated across 11 systems

• 20+ AI agents collaborating autonomously with humans in the loop

• End-to-end process time reduced from weeks to minutes

Purple Fabric's innovative complaints resolution framework transformed how investigations are managed—unlocking speed, quality, and scale while minimizing compliance risks.

Figure 4. Behind the scenes, the agent team connects with 11+ systems retrieving customer details, transaction history, policy documents, and regulatory guidelines — to build a complete contextual foundation for the complaint.

Figure 5. Once the case dossier is compiled, the final layer of agents - the investigation agent and report writing agent scrutinises the information, produces an investigation report, and recommends a resolution — complete with supporting evidence.

Figure 6. Every decision, action, and agent interaction is logged in TRACES — Purple Fabric's explainability layer, so that you can follow exactly how the recommendation was reached.

Figure 7. Every insight is backed by evidence. 'Sources' reveals which documents were used and how, providing full traceability from complaint to conclusion.

Figure 8. Each agent in the Complaints Investigator team is modular and API-publishable. This means the entire workflow can be embedded into existing enterprise applications, enabling intelligent complaint resolution without disrupting current systems.

Use Case: Corporate Intelligence

From 100,000 person-days to intelligent ESG insights.

A leading institutional investor managing a portfolio of over 9,000 companies faced a daunting challenge: integrating ESG (Environmental, Social, and Governance) insights into portfolio analysis. ESG data was highly fragmented across thousands of sources—including reports, disclosures, filings, and news articles—resulting in an overwhelming 10 million documents for analysts to sift through manually.

Portfolios shifted by nearly 10% each month, ESG ratings proved inconsistent, and mounting regulatory pressure demanded sustainability alignment. Analysts spent an estimated 100,000 person-days annually reviewing documents, piecing together insights, and generating reports—an effort that was both time-intensive and error-prone.

Purple Fabric transformed this process by deploying a multi-agent Corporate Intelligence system designed to automate and elevate ESG analysis.

The solution began with agents that ingested ESG documents across structured and unstructured formats, creating a unified ESG knowledge garden. Then, specialised digital experts extracted and aligned data to client-specific ESG guidelines ensuring that every insight mapped to the investor's frameworks.

Agents like the Environment Agent, Social Data Agent, and Governance Data Agent parsed sustainability-related information at scale. A 360 Analyst Agent created holistic company profiles. A Query Agent enabled users to interrogate the entire portfolio using natural language, and a Reporting Agent generated investor-ready summaries which were all traceable and audit-ready.

To keep the intelligence real-time, APIs were used to ingest news and external data feeds, enabling agents to perform sentiment analysis on the latest ESG developments. This allowed the platform to identify emerging risks and opportunities across companies, ahead of manual review cycles.

Every insight was grounded in original data with explainability by design. And since agents operated with real-time access to both historical records and live APIs, the analysis was always current.

The results?

• Near 100% portfolio coverage

• Over 90% accuracy in extracted ESG records

• 1,000× faster turnaround times compared to traditional methods

• 100,000 specialist person-days saved annually

What once required brute-force manual review now runs on intelligent, collaborative agents, delivering ESG analysis with unprecedented speed, scale, and traceability.

Figure 9. The most comprehensive and differentiated knowledge base for ESG and non-financial portfolio analysis — built from over 10 million documents across 9,000 companies and structured into 60 billion+ searchable chunks.

Figure 10. Natural language queries are answered using company-specific documents. Each response is explainable through TRACES and links to the underlying source material.

Figure 11. Users get a full 360° company view in seconds, curated by specialised ESG agents and presented through a structured interface with differentiated, actionable insights for portfolio management and ESG reporting.

Figure 12. Users can explore insights across multiple levels — from portfolio dashboards to individual company scores, all the way down to the original source documents that power every datapoint.

Deploy within your own VPC

Enterprises can host Purple Fabric entirely within their own cloud environment, typically on AWS or Azure, enabling them to maintain full control over data, access, and infrastructure. In this model, all enterprise knowledge, agent orchestration, and governance components run inside the client's own Virtual Private Cloud (VPC), adhering to their internal security, audit, and compliance policies.

Dedicated Tenant on Intellect's Cloud

For enterprises seeking faster deployment without infrastructure overhead, Purple Fabric can be provisioned as a dedicated tenant on Intellect's secure cloud. Enterprises connect their internal systems via private, credential-based connections that ensure data access remains secure and tightly governed. Each tenant is logically isolated with independent governance, observability, and compute controls, ensuring enterprise-grade security while accelerating time-to-production.

Regardless of the deployment path, Purple Fabric offers:

• Built-in monitoring and observability: Including real-time dashboards, cost tracking, agent execution trails, and system health metrics

• Secure access to enterprise systems: Managed via the Credentials Module, which stores and encrypts API keys, database credentials, and email connector tokens for seamless and secure integration

• Security and access control: Granular, role-based access and workspace-level isolation

• Auditability and compliance: Every document retrieval, agent action, and LLM output is logged for end-to-end traceability, with support for encryption at rest and in transit

Whether hosted within an enterprise's cloud environment or managed as a secure, isolated tenant in Intellect's cloud, Purple Fabric ensures that AI is deployed securely, monitored transparently, and governed at every step.

Next Steps

Learn more about Purple Fabric and connect with the IntellectAI team at purplefabric@intellectdesign.com. Learn more about MongoDB Atlas and its enterprise-grade database solutions.