Trusted, Cited AI Answers from Corporate Data, with Permissions Enforced
A permission-aware RAG platform that grounds every AI answer in the right company data
About the customer
Our client is a US generative AI company that helps enterprises use AI with their own corporate data safely. Employees want AI assistants that answer from company documents, spreadsheets and databases, not just public knowledge. The client provides a trusted, permission-aware Retrieval Augmented Generation (RAG) platform, which grounds every AI answer in the right corporate data, for the right user, with full governance.
What needed solving
Scattered knowledge
Enterprise knowledge was spread across Google Drive, SharePoint, OneDrive, S3 and data warehouses like Snowflake, out of reach of generic AI assistants
Many formats
Data came in many formats, including scanned PDFs, slides and spreadsheets
Risk of exposure
Connecting this data naively risked exposing confidential files to employees who should not see them
Accurate, cited answers
The solution had to respect access permissions and sensitivity labels while returning accurate, cited answers across chat, search and AI agents
How we built it
Data sync
Syncs corporate data from Google Drive, SharePoint, OneDrive, Amazon S3, MinIO, Snowflake and direct file uploads, with incremental sync (only changes are updated), scheduled jobs, and pause, resume and retry controls.
Parsing and indexing
OCR (text recognition for scanned files) and optional vision-language models (VLMs, AI that reads images) parse complex PDFs, tables and images. Documents are split into chunks and indexed in Qdrant or Azure AI Search.
Hybrid retrieval
Combines meaning-based and keyword-based (BM25) search, merges the rankings and reranks the results. An agentic retrieval agent also blends vector search with text-to-SQL, which turns questions into database queries, for structured data.
Governance
Every result carries citations and is filtered per user using identity provider permissions. Admins include or exclude data by sensitivity label, and ingested content is screened for malicious or poisoned chunks.
Product experiences
Powers trusted answers in chat, enterprise search, document chat and document summary, plus knowledge bases for AI agents.
Models and deployment
Supports OpenAI, Azure OpenAI, AWS Bedrock, Gemini, Mistral and self-hosted open-source models for fully on-premises deployments. Moved to production once results were stable.
Corporate knowledge behind AI assistants and agents, without compromising data security
Technologies
Domains
Have a Query? Let's Have a Conversation
Do you have a project in mind? We’d love to know what it is. Email us or send a message from our contact page.
Have a similar problem?
Tell us what you are trying to solve and our team will get back to you.