IntelDocs AI

Project Story

About the Project

I built IntelDocs AI to solve a common workplace problem: valuable knowledge is scattered across documents, making it difficult for teams to find accurate information quickly. The goal was to create a centralized knowledge platform where employees can ask questions in natural language while keeping company-wide and team-specific information securely separated.

IntelDocs AI combines document management, company and team workspaces, dashboards, and AI-powered chat in one platform. Its multi-tenant architecture allows organizations to manage their knowledge centrally while ensuring that team documents and conversations remain restricted to the appropriate team.

How It Works

I built the backend using FastAPI and asynchronous SQLAlchemy, with PostgreSQL and pgvector for structured data storage and semantic search. The RAG pipeline processes supported document formats, splits their content into meaningful chunks, generates embeddings using a local Hugging Face model, and retrieves relevant passages to ground responses generated through Groq. The frontend uses HTML, CSS, and JavaScript to provide interactive company and team dashboards.

Challenges and Learnings

One of the biggest lessons was that reliable RAG involves much more than connecting a language model to a prompt. Document parsing, chunking strategies, retrieval quality, and source context all play a critical role in generating useful answers.

I also designed access-control and retrieval flows around company-level and team-level permissions, ensuring that search results respect the user's access scope. To improve responsiveness, I addressed blocking model operations in asynchronous API workflows and initialized the embedding model at application startup to avoid unnecessary loading delays on the first request.

Building IntelDocs AI strengthened my understanding of RAG system design, semantic search, multi-tenant architecture, asynchronous backend development, and access-controlled AI applications.

Built With

• ci/cd
• css
• docker
• fastapi
• html
• javascript
• langchain
• postgreesql
• pytest
• python
• rag
• supabase