DS
Diwesh Saxena
70% ↓ search time

Enterprise RAG: internal knowledge base for 200+ consultants

Fractional CTOManagement consulting firmAug 2023 — Feb 2024Published 22 October 2023

Built a retrieval-augmented knowledge platform indexing proposals, playbooks, and case archives — with citation grounding and access controls.

Technology Stack

PythonOpenAIPineconeReactFastAPIAzure

Key Outcomes

  • 70% reduction in time spent searching historical proposals and decks
  • Citation-grounded answers with source links for auditability
  • Role-based access preserving client confidentiality
Outcome: 70% less time searching proposals and institutional knowledge.

Context

A 200+ person consulting firm stored a decade of proposals across SharePoint, email, and local drives. Partners needed answers fast without breaching client confidentiality.

Problem

  • Search tools returned filenames, not answers.
  • Junior staff recreated slides that already existed.
  • No guardrails against cross-client data leakage.

Approach

  1. Ingestion — PDF/PPTX parsers with client-level tagging.
  2. Chunking + embeddings — Pinecone with metadata filters.
  3. RAG answers — mandatory citations; “I don’t know” when confidence low.
  4. RBAC — Azure AD groups map to index partitions.

Results

Self-reported search time down 70% in a 30-day pilot; legal approved rollout after DPIA review.

Lessons

Citations are not optional in professional services. RBAC at the vector layer beats post-hoc filtering.

CTA

Need enterprise RAG with real access controls? Book a 30-min call.