IAA-RAG v2
The Most Advanced AI Research Platform
Built for Space Science
A state-of-the-art Retrieval-Augmented Generation system engineered from the ground up at the Institute of Space Science — combining cutting-edge AI models, hybrid search, knowledge graphs and real-time document intelligence.
Technology Stack
Gemini 2.0 Flash
Google's latest LLM for generation, vision analysis and cross-document reasoning. LaTeX, images and academic citations fully supported.
Hybrid Search Engine
Semantic vector search (768-dim embeddings via gemini-embedding-001) combined with BM25 lexical search and re-ranking for top-35 precision retrieval.
Knowledge Graph (Neo4j)
3,300+ entities and 1,600+ relations extracted from documents. Powers contradiction detection, cross-document analysis and agentic reasoning.
Real-Time Streaming
FastAPI streaming responses with Server-Sent Events. Answers appear token-by-token, no waiting. Parallel worker processing via Redis queues.
Multi-Format Ingestion
PDF, DOCX, PPTX, XLSX, images (AI vision analysis), YouTube transcripts, web scraping — all automatically chunked, embedded and indexed.
Agentic Deep Research
Multi-step reasoning engine that decomposes complex questions into sub-queries, researches each independently, then synthesizes a comprehensive answer.
Auto Report Generation
One click generates a full academic report — executive summary, structured sections, citations — exported as HTML ready for publication.
Production Infrastructure
Fully containerized: FastAPI · PostgreSQL + pgvector · Qdrant · Neo4j · Redis · MinIO · Nginx · Prometheus/Grafana monitoring. 4 parallel worker replicas.
Academic Citation System
Every answer includes clickable inline citations [1] linked to the exact page or timestamp in the source document. Authors, journal, year and page number — automatically extracted. Click a citation to jump directly to the source.
YouTube & Multimedia Intelligence
Ingest any YouTube video — transcripts are automatically extracted, chunked and indexed. Citations link to the exact timestamp in the video. Images in documents are analyzed with AI vision and included in search.
Contradiction Detection
The Knowledge Graph engine identifies when documents contradict each other on the same topic. Unique capability for research quality assurance — compare positions across years and authors automatically.
n8n Webhook Integration
Full automation API: trigger document ingestion, query the RAG engine or generate reports from any external system via webhooks. Integrates with n8n, Zapier, Make and custom pipelines out of the box.
Platform Screenshots
Built at the Institute of Space Science · Powered by Google Gemini 2.0