Turn fragmented knowledge into a
connected intelligence layer.
A cloud-based GraphRAG knowledge engine that combines knowledge graphs, vector retrieval, and document intelligence to connect information scattered across educational documents.
Why conventional search fails
Traditional document search treats educational materials as isolated chunks of text. It loses the most important part of learning: the connections between concepts.
Disconnected Documents
Information is scattered across PDFs, lecture notes, and assignments without cross-referencing capabilities.
Fragmented Concepts
Keyword search retrieves isolated sentences, missing the broader context necessary for deep understanding.
Hidden Relationships
Unable to connect ideas across different lectures or pinpoint how foundational concepts map to advanced topics.
Messy Data
Struggles with unstructured whiteboard screenshots, handwritten notes, and informal study material.
The Knowledge Pipeline
From fragmented files to a connected intelligence layer in seconds.
Multimodal Ingestion
Upload PDFs, lecture notes, PowerPoints, and whiteboard screenshots.
Semantic Extraction
Entity and relationship detection from raw text and OCR data.
Knowledge Graph
Construct a living web of connected concepts alongside vector retrieval.
GraphRAG Reasoning
Deliver source-grounded answers by traversing the graph and retrieving relevant vectors.
Vector + Graph Architecture
AetherGraph doesn't just search text; it traverses relationships. By combining HNSW ANN vector search for semantic retrieval with a PostgreSQL-backed knowledge graph, it grounds answers in actual structural connections.
Built for messy educational data
Real-world study materials aren't clean text files. AetherGraph is designed to handle the reality of student data, parsing through noisy inputs to extract meaningful connections.
- ✓PDFs & Lecture Notes: Extracts structured text and metadata from extensive academic documents.
- ✓PowerPoints: Parses slide content to maintain the sequence of concepts.
- ✓Whiteboard Screenshots: Processes images of handwritten notes and diagrams (best-effort extraction).
- ✓Assignments & Syllabi: Automatically extracts deadlines and topics into an interactive format.
Engineering & Infrastructure
AetherGraph is a serious engineering project built with an emphasis on modular architecture, cloud deployment, and scalable data processing. It is deployed across AWS, Render, and Vercel.
Multi-Cloud Deployment
Orchestrated AWS, Render, and Vercel into a secure multi-cloud architecture with isolated networking. Dockerized backend with a dedicated pgvector PostgreSQL container on AWS.
HNSW ANN Vector Search
Engineered Hierarchical Navigable Small World (HNSW) Approximate Nearest Neighbor search for fast, low-latency semantic retrieval across thousands of pages.
Asynchronous FastAPI Backend
Deployed a production-ready asynchronous FastAPI backend on Render. Handled async connection pools and kept the system within strict memory limits by offloading inference.
Real-time Synchronization
Added real-time WebSocket synchronization between the Next.js frontend and the FastAPI backend for live GraphRAG processing updates.
Technology Stack
The Project
AetherGraph is built independently by a student engineer exploring practical AI infrastructure and knowledge systems. It was designed to push the boundaries of standard web apps, turning fragmented data into a living web of connected concepts that students can actually explore and learn from.
Early-stage engineering project under active development.