OmniSearch AI Assistant
A custom semantic search tool parsing corporate manual files to answer agent inquiries in real-time, built using vector embeddings.
The Challenge
The client's support agents were spending an average of 8 minutes per ticket digging through hundreds of pages of technical PDFs and operation manuals. This led to high customer wait times, agent fatigue, and low resolution rates on the first contact.
The Solution
We engineered OmniSearch, a semantic AI assistant. It ingests all corporate manuals, chunks the text, and stores high-dimensional embeddings in a Pinecone vector database. When agents ask a natural language question, the system queries the vector space to find relevant context, then synthesizes a precise answer using the Gemini API. The interface is built on a fast Next.js React frontend.
Technical Architecture
# Pipeline & Framework Specs
Next.js (App Router) client app connecting to a Python/FastAPI backend, utilizing SentenceTransformers for text embeddings, Pinecone for vector similarity index search, and Google Gemini API for context-based generation.
