KgpOne
KGPOne is not a generic SaaS management platform. It is a reusable education + AI engine that powers different institutional builds. The engine is shared. Institution-specific workflows, terminology, and UI are custom code built on top of the engine not configuration flags.
Key Features
- Role-Based Access Control (RBAC): Secure endpoints protected by JWT authentication with rotating refresh tokens, scoping access across STUDENT, TA, PROFESSOR, and ADMIN roles.
- Metadata-Driven Vector Search: Highly optimized semantic search queries executed directly on Qdrant, filtering securely using pre-indexed payload schemas mapped from PostgreSQL.
- Modular Background Ingestion: Uploaded PDFs are stored in S3, and asynchronously processed by Celery. The pipeline splits tables, maths, and texts, embeds them, and persists them into Qdrant.
Tech Stack
My Contribution
Designed the REST APIs, PostgreSQL schema, authentication flow, Redis caching, and Docker deployment.
Technical Highlight
Designed the system around PostgreSQL transactions to maintain data consistency.
Architecture & Workflow
The architecture is deliberately layered: A **document intelligence layer** that deeply understands structure, A **knowledge layer** that extracts entities and relationships, A **retrieval layer** that fuses vector similarity with graph traversal, An **analysis layer** that uses LLMs to synthesize study materials.