Enterprise Knowledge Search · Security and Compliance
Intellect Design Arena needed a governed retrieval architecture that could turn fragmented ESG data into trustworthy enterprise AI answers for highly regulated financial institutions.
Reported impact
The Purple Fabric architecture improved ESG answer accuracy, cut P95 latency roughly in half, and moved from implementation to production in six to eight weeks while maintaining governance for ambiguous queries.
Intellect Design Arena built its Purple Fabric enterprise AI platform around MongoDB Atlas to operationalize governed AI for regulated finance use cases such as ESG analysis for a sovereign wealth fund. Source material arrives as PDFs, images, and public web documents rather than neat tabular datasets, so the platform had to preserve lineage and relevance while scaling beyond selective manual reviews. Purple Fabric separates data gathering from application and reasoning layers, then uses MongoDB Atlas as the always-available data foundation for a unified retrieval pipeline. Metadata filters are applied early to narrow large corpora before hybrid retrieval combines vector search, lexical search, and Reciprocal Rank Fusion reranking. The architecture evaluates retrieval, relevance, and answer quality as distinct stages, and routes the hardest remaining cases into human-in-the-loop review. This lets Intellect improve answer precision and governance without relying on model improvements alone.
PDFs, images, and public web documents are ingested into Purple Fabric and stored on MongoDB Atlas; metadata filters narrow the corpus, hybrid lexical and vector retrieval with RRF ranks evidence, and low-confidence cases are escalated to human reviewers before answers are exposed to end users.
P95 latency was also reduced by approximately half.
With hybrid search and Reciprocal Rank Fusion (RRF), accuracy increased by 17%.
“For anyone who wants to productionize AI, focus more on architecture, focus more on engineering. Models will continue to get better, but that will not help you to move to production.”
“Governance cannot be an afterthought. You have to enforce it in the pipeline so that it’s taken care of before you expose anything to the LLM.”
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Google Cloud Sample Archived · Apr 22, 2026 PayPal PayPal wanted thousands of employees to query petabytes of governed finance data conversationally, but it had to avoid plaintext secrets, AI hallucinations, inconsistent metric definitions, and data-loss-prevention… Achieved near 100% accuracy in AI-generated insights BigQuery · Claude · Looker 
OpenAI Sample Dec 12, 2025 BNY BNY wanted to scale generative AI across a systemically important financial institution without treating it as a side experiment, so it needed a governed way for employees to build agents, use frontier models, and… 20,000 employees actively building AI agents ChatGPT Enterprise · Eliza · deep research 
Microsoft Sample Dec 11, 2024 Crediclub Crediclub recognized the need to enhance the operational management of its over 150 branches to maintain the consistency and quality of its processes. Crediclub cut monthly costs by 96% Azure AI Services · Azure Cosmos DB · Azure OpenAI Service 
