HiveBrief

Problem

Teams drown in documents — PDFs, reports, specs — but can't quickly find answers. Search fails because the answer often spans multiple documents and requires synthesis.

Approach

HiveBrief ingests documents (PDF, DOCX), creates vector embeddings with pgvector, and uses Gemini 2.5 Flash for RAG-based question answering with source citations. Every answer links back to the specific passages it draws from.

Key Metrics

• Sub-second retrieval on document sets up to 10K pages
• Source citation accuracy verified against ground truth
• Deployed on AWS EC2 and RDS behind a Cloudflare tunnel

Tech Stack Detail

Next.js frontend, FastAPI backend with PostgreSQL + pgvector for embeddings, Gemini 2.5 Flash for generation, AWS RDS for database, S3 for document storage. RAG pipeline with chunk-level citation tracking.