RAG / Internal AI products

Employee Knowledge & CRM

Making company procedures searchable and turning call summaries into useful customer records.

GnG Vacation · Sole developer

Hybrid

vector, keyword and title retrieval

The problem

New employees needed a clear way to find procedures, responsibilities and training materials. Sales staff also needed customer conversations to carry through into CRM records. I built both workflows around the company’s internal knowledge and operations tools.

How it works

  1. 01

    Documents

    Training materials and company procedures

  2. 02

    Index

    Chunking, embeddings and pgvector

  3. 03

    Retrieve

    Vector + keyword + title search, merged with RRF

  4. 04

    Answer

    An assistant that can search and read articles

What I built

  • A document ingestion and retrieval path with chunking, embeddings, a pgvector index and article-reading tools.
  • Hybrid search combining semantic similarity, keywords and title matches with reciprocal rank fusion.
  • Employee access to learning materials and an AI assistant for questions during day-to-day work.
  • Zoom call-summary synchronization, AI-assisted call classification and customer-record updates in the CRM.

Engineering decisions

Use more than semantic similarity

Exact policy names and team terminology matter. Keyword and title matches complement vector retrieval, and reciprocal rank fusion combines the result lists.

Make customer-record changes reviewable

Confidence checks gate lead creation, while audit records preserve the classification and update decisions for later inspection.

Keep the source granularity clear

The CRM integration processes Zoom call summaries. Its description does not assume that every call has a complete verbatim transcript.

View résumés [email protected]