AssistOS

Overview

AssistOS is an end-to-end Retrieval-Augmented Generation platform that enables document ingestion, semantic search, and streaming AI responses with transparent source citations. Built as a cloud-native system for accurate, grounded answers over private document corpora.

Problem

Teams need reliable AI answers grounded in their own documents—not generic LLM hallucinations. That requires a complete pipeline: ingestion, chunking, embeddings, vector retrieval, and streaming generation with citations.

Solution

Designed and built a full-stack RAG system with a Next.js frontend, FastAPI backend, Gemini embeddings, and Qdrant vector storage. Document chunking and retrieval are optimized for complex queries, while responses stream with source citations for trust and auditability.

Architecture

Technical decisions

• Chose Qdrant for efficient vector similarity search and retrieval accuracy on complex queries.
• Used Gemini for embeddings to balance quality and integration simplicity in the RAG pipeline.
• Containerized backend services with Docker for consistent environments and portable deployments.
• Deployed frontend and API on Vercel/Render to keep the architecture stateless and horizontally scalable.
• Streamed responses so users see progressive output while retrieval and generation complete.

Key highlights

• Document ingestion and semantic search
• Streaming AI responses with source citations
• Document chunking and vector embeddings via Gemini + Qdrant
• Dockerized services with Vercel/Render deployment
• Stateless, scalable architecture